@devries-cameron20 / t3code Mirror · read-only / commits / 55a94e2029
Fork 0

The best way to control Claude Code, Codex and more from any computer or phone

Mirrored from https://github.com/Camo67/t3code.git

Synced Nov 21, 21:00 UTC

fix(web): voice settings and dictation match the rest of the app (#33140)

maria committed Nov 12, 2026 at 04:10 UTC 55a94e20293ff447ecf3b21e0d93a87abb0a2344
26 files changed +1495 -1107
apps/web/src/components/chat/AssistantCitationCommentEditor.tsx
+1 -9
@@ -113,15 +113,7 @@ export function AssistantCitationCommentEditor({
113 </p>
114 ) : null}
115 <div className="mt-2 flex flex-wrap items-center justify-end gap-2">
116 - <div
117 - className={
118 - speech.state.phase === "idle"
119 - ? "contents"
120 - : "flex basis-full flex-wrap gap-2 *:min-w-0 *:grow **:data-[slot=voice-input-pill]:w-full"
121 - }
122 - >
123 - <TextFieldSpeechControls speech={speech} />
124 - </div>
116 + <TextFieldSpeechControls speech={speech} />
117 <Button
118 variant="outline"
119 size="xs"
apps/web/src/components/chat/ChatComposer.tsx
+17 -11
@@ -985,6 +985,7 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
985 null,
986 );
987 const [isDragOverComposer, setIsDragOverComposer] = useState(false);
988 + const [isComposerFooterCompact, setIsComposerFooterCompact] = useState(false);
989 const [isComposerPrimaryActionsCompact, setIsComposerPrimaryActionsCompact] = useState(false);
990 const [isComposerModelPickerOpen, setIsComposerModelPickerOpen] = useState(false);
991 const isMobileViewport = useMediaQuery("max-sm");
@@ -1484,19 +1485,18 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
1485 useLayoutEffect(() => {
1486 const composerForm = composerFormRef.current;
1487 if (!composerForm) return;
1487 - const measurePrimaryActionsCompact = () => {
1488 + const measureCompactLayouts = () => {
1489 const composerFormWidth = composerForm.clientWidth;
1490 const options = { hasWideActions: composerFooterHasWideActions };
1490 - return (
1491 + setIsComposerFooterCompact(shouldUseCompactComposerFooter(composerFormWidth));
1492 + setIsComposerPrimaryActionsCompact(
1493 shouldUseCompactComposerFooter(composerFormWidth, options) &&
1492 - shouldUseCompactComposerPrimaryActions(composerFormWidth, options)
1494 + shouldUseCompactComposerPrimaryActions(composerFormWidth, options),
1495 );
1496 };
1497
1496 - setIsComposerPrimaryActionsCompact(measurePrimaryActionsCompact());
1497 - return observeResize(composerForm, () => {
1498 - setIsComposerPrimaryActionsCompact(measurePrimaryActionsCompact());
1499 - });
1498 + measureCompactLayouts();
1499 + return observeResize(composerForm, measureCompactLayouts);
1500 }, [composerFooterHasWideActions]);
1501
1502 // oxlint-disable-next-line t3code/no-unreasoned-state-sync -- reads draft images as data URLs and persists them to the draft store, dropping stale results on cleanup
@@ -1825,6 +1825,8 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
1825 },
1826 });
1827 const { blocksSubmission: voiceInputActive, submitAfterDictation } = speechInput;
1828 + // A compact footer gives the dictation capsule the whole row beside the send button.
1829 + const dictationTakesFooter = isComposerFooterCompact && speechInput.state.phase !== "idle";
1830 const submissionTarget = activePendingProgress ? "pending-user-input" : "provider-turn";
1831 const submissionBlocked = isComposerSubmissionBlocked({
1832 submissionTarget,
@@ -3581,7 +3583,9 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
3583 "hidden sm:flex",
3584 )}
3585 >
3584 - {composerAttachButton ? <div className="me-2">{composerAttachButton}</div> : null}
3586 + {composerAttachButton && !dictationTakesFooter ? (
3587 + <div className="me-2">{composerAttachButton}</div>
3588 + ) : null}
3589 <div
3590 data-chat-composer-actions="right"
3591 data-chat-composer-transition-actions="true"
@@ -3594,7 +3598,10 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
3598 ref={attachExpandedComposerControls}
3599 data-chat-composer-controls="footer"
3600 data-chat-composer-footer-controls="true"
3597 - className="relative flex min-w-0 flex-1 items-center justify-end gap-0.5"
3601 + className={cn(
3602 + "relative flex min-w-0 flex-1 items-center justify-end gap-0.5",
3603 + dictationTakesFooter && "hidden",
3604 + )}
3605 >
3606 {composerControls}
3607 </div>
@@ -3638,6 +3645,7 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
3645 onPreviousPendingQuestion={onPreviousActivePendingUserInputQuestion}
3646 onInterrupt={handleInterruptPrimaryActionStable}
3647 compactBeforeSendTokens={props.resumeCompactionTokens}
3648 + hideCompactToggle={dictationTakesFooter}
3649 keepFullHistory={props.keepFullHistory}
3650 onToggleKeepFullHistory={props.onToggleKeepFullHistory}
3651 >
@@ -3657,11 +3665,9 @@ export const ChatComposer = memo(function ChatComposer(props: ChatComposerProps)
3665 />
3666 <ComposerSpeechButton
3667 state={speechInput.state}
3660 - progress={null}
3668 shortcutLabel={dictationShortcutLabel}
3669 disabled={dictationDisabled}
3670 onStart={() => void speechInput.start()}
3664 - onCancel={() => speechInput.cancel()}
3671 />
3672 </>
3673 ) : null}
apps/web/src/components/chat/ComposerFooterControls.tsx
+2
@@ -216,6 +216,7 @@ export const ComposerFooterPrimaryActions = memo(function ComposerFooterPrimaryA
216 onPreviousPendingQuestion: () => void;
217 onInterrupt: () => void;
218 compactBeforeSendTokens: number | null;
219 + hideCompactToggle: boolean;
220 keepFullHistory: boolean;
221 onToggleKeepFullHistory: () => void;
222 }) {
@@ -245,6 +246,7 @@ export const ComposerFooterPrimaryActions = memo(function ComposerFooterPrimaryA
246 onPreviousPendingQuestion={props.onPreviousPendingQuestion}
247 onInterrupt={props.onInterrupt}
248 compactBeforeSendTokens={props.compactBeforeSendTokens}
249 + hideCompactToggle={props.hideCompactToggle}
250 keepFullHistory={props.keepFullHistory}
251 onToggleKeepFullHistory={props.onToggleKeepFullHistory}
252 >
apps/web/src/components/chat/ComposerPrimaryActions.tsx
+3 -1
@@ -62,6 +62,7 @@ interface ComposerPrimaryActionsProps {
62 onPreviousPendingQuestion: () => void;
63 onInterrupt: () => void;
64 compactBeforeSendTokens?: number | null;
65 + hideCompactToggle?: boolean;
66 keepFullHistory?: boolean;
67 onToggleKeepFullHistory?: () => void;
68 }
@@ -118,6 +119,7 @@ export const ComposerPrimaryActions = memo(function ComposerPrimaryActions({
119 onPreviousPendingQuestion,
120 onInterrupt,
121 compactBeforeSendTokens = null,
122 + hideCompactToggle = false,
123 keepFullHistory = false,
124 onToggleKeepFullHistory,
125 }: ComposerPrimaryActionsProps) {
@@ -356,7 +358,7 @@ export const ComposerPrimaryActions = memo(function ComposerPrimaryActions({
358 <TooltipPopup>{submitTooltip}</TooltipPopup>
359 </Tooltip>
360 );
359 - if (compactTokens === null) {
361 + if (compactTokens === null || hideCompactToggle) {
362 return withStopButton(
363 <>
364 {children}
apps/web/src/components/chat/ComposerSpeechButton.tsx
+111 -191
@@ -1,6 +1,6 @@
1 import type { VoiceInputState } from "@t3tools/client-runtime/voice-input";
2 import { CheckIcon, MicIcon, XIcon } from "lucide-react";
3 -import { useEffect, useState } from "react";
3 +import { useEffect, useState, type PointerEvent, type ReactElement } from "react";
4
5 import { VoiceInputErrorPill } from "./VoiceInputErrorPill";
6 import { VoiceInputPill, VoiceWaveform } from "./VoiceInputPill";
@@ -9,69 +9,33 @@ import { Button } from "../ui/button";
9 import { Spinner } from "../ui/spinner";
10 import { Tooltip, TooltipPopup, TooltipTrigger } from "../ui/tooltip";
11
12 -type SpeechPresentation = {
13 - status: string | null;
14 - showsCancel: boolean;
15 - showsConfirm: boolean;
16 - confirmEnabled: boolean;
17 - showsSend: boolean;
18 -};
12 +type DownloadProgress = { downloaded: number; total: number } | null;
13
14 +const keepFocus = (event: PointerEvent) => event.preventDefault();
15 +
16 +/** One short status line per phase; also mirrored into the preview annotation editor. */
17 export function resolveSpeechPresentation(
18 state: VoiceInputState,
22 - progress: { downloaded: number; total: number } | null,
23 -): SpeechPresentation {
19 + progress: DownloadProgress,
20 +): { status: string | null; active: boolean } {
21 switch (state.phase) {
22 case "idle":
26 - return {
27 - status: null,
28 - showsCancel: false,
29 - showsConfirm: false,
30 - confirmEnabled: false,
31 - showsSend: true,
32 - };
23 + return { status: null, active: false };
24 case "error":
34 - return {
35 - status: state.error ?? "Voice input failed",
36 - showsCancel: false,
37 - showsConfirm: false,
38 - confirmEnabled: false,
39 - showsSend: true,
40 - };
25 + return { status: state.error ?? "Voice input failed", active: false };
26 case "preparing":
27 return {
28 status: progress
44 - ? `Downloading speech model ${Math.round((progress.downloaded / Math.max(1, progress.total)) * 100)}%`
45 - : "Preparing",
46 - showsCancel: true,
47 - showsConfirm: true,
48 - confirmEnabled: false,
49 - showsSend: true,
29 + ? `Downloading model ${Math.round((progress.downloaded / Math.max(1, progress.total)) * 100)}%`
30 + : "Starting…",
31 + active: true,
32 };
33 case "recording":
52 - return {
53 - status: "Recording",
54 - showsCancel: true,
55 - showsConfirm: true,
56 - confirmEnabled: true,
57 - showsSend: true,
58 - };
34 + return { status: "Recording", active: true };
35 case "transcribing":
60 - return {
61 - status: "Transcribing",
62 - showsCancel: true,
63 - showsConfirm: true,
64 - confirmEnabled: false,
65 - showsSend: true,
66 - };
36 + return { status: "Transcribing…", active: true };
37 case "post-processing":
68 - return {
69 - status: "Post-processing",
70 - showsCancel: true,
71 - showsConfirm: true,
72 - confirmEnabled: false,
73 - showsSend: true,
74 - };
38 + return { status: "Cleaning up…", active: true };
39 }
40 }
41
@@ -94,81 +58,31 @@ function RecordingStatus(props: { level: number; recordingLimitSeconds?: number
58
59 return (
60 <>
97 - <VoiceWaveform level={props.level} />
98 - <span aria-hidden className="size-1.5 shrink-0 rounded-full bg-primary" />
99 - <span className="shrink-0 text-secondary-label text-xs tabular-nums">
61 + <span aria-hidden className="size-2 shrink-0 rounded-full bg-destructive" />
62 + <span className="shrink-0 text-xs text-muted-foreground tabular-nums">
63 {formatElapsed(elapsedSeconds)}
64 {props.recordingLimitSeconds !== undefined
65 ? ` / ${formatElapsed(props.recordingLimitSeconds)}`
66 : ""}
67 </span>
68 + <div className="me-1.5 flex w-16 min-w-0 shrink grow">
69 + <VoiceWaveform level={props.level} />
70 + </div>
71 </>
72 );
73 }
74
109 -function ComposerSpeechStatus(props: {
110 - state: VoiceInputState;
111 - progress: { downloaded: number; total: number } | null;
112 - level: number;
113 - recordingLimitSeconds?: number | undefined;
114 -}) {
115 - const presentation = resolveSpeechPresentation(props.state, props.progress);
116 -
117 - if (!presentation.status) return null;
118 - const isRecording = props.state.phase === "recording";
119 - const isError = props.state.phase === "error";
120 -
121 - return (
122 - <div
123 - className="flex h-7 min-w-0 flex-1 items-center gap-2"
124 - role="status"
125 - aria-live={isRecording ? "off" : "polite"}
126 - aria-label={presentation.status}
127 - >
128 - {isRecording ? (
129 - <RecordingStatus level={props.level} recordingLimitSeconds={props.recordingLimitSeconds} />
130 - ) : (
131 - <span
132 - className={cn(
133 - "min-w-0 flex-1 truncate text-right text-sm text-secondary-label",
134 - isError && "text-destructive",
135 - )}
136 - >
137 - {presentation.status}
138 - </span>
139 - )}
140 - </div>
141 - );
142 -}
143 -
144 -function ComposerSpeechCancelButton(props: {
145 - state: VoiceInputState;
146 - shortcutLabel?: string | null;
147 - onCancel(): void;
75 +function ShortcutTooltip(props: {
76 + label: string;
77 + shortcut?: string | null | undefined;
78 + children: ReactElement;
79 }) {
149 - if (props.state.phase === "idle") return null;
150 - const label = props.state.phase === "error" ? "Dismiss voice input error" : "Cancel voice input";
151 -
80 return (
81 <Tooltip>
154 - <TooltipTrigger
155 - render={
156 - <Button
157 - type="button"
158 - size="icon-sm-pill"
159 - variant="ghost"
160 - aria-label={label}
161 - onPointerDown={(event) => event.preventDefault()}
162 - onClick={props.onCancel}
163 - className="shrink-0"
164 - >
165 - <XIcon />
166 - </Button>
167 - }
168 - />
82 + <TooltipTrigger render={props.children} />
83 <TooltipPopup side="top">
170 - {label}
171 - {props.shortcutLabel ? ` (${props.shortcutLabel})` : ""}
84 + {props.label}
85 + {props.shortcut ? ` (${props.shortcut})` : ""}
86 </TooltipPopup>
87 </Tooltip>
88 );
@@ -176,63 +90,56 @@ function ComposerSpeechCancelButton(props: {
90
91 export function ComposerSpeechButton(props: {
92 state: VoiceInputState;
179 - progress: { downloaded: number; total: number } | null;
93 shortcutLabel?: string | null;
94 disabled?: boolean;
95 + /** Compact matches the `xs` buttons beside text-field editors. */
96 + size?: "default" | "compact";
97 onStart(): void;
183 - onCancel(): void;
98 }) {
185 - const presentation = resolveSpeechPresentation(props.state, props.progress);
99 + if (props.state.phase !== "idle") return null;
100 const label = "Start voice input";
101
188 - if (presentation.showsConfirm || props.state.phase === "error") return null;
189 -
102 return (
191 - <Tooltip>
192 - <TooltipTrigger
193 - render={
194 - <Button
195 - type="button"
196 - size="icon-sm"
197 - variant="ghost"
198 - aria-label={label}
199 - disabled={props.disabled}
200 - onPointerDown={(event) => event.preventDefault()}
201 - onClick={() => {
202 - if (props.disabled) return;
203 - props.onStart();
204 - }}
205 - className="relative shrink-0"
206 - >
207 - <MicIcon />
208 - </Button>
209 - }
210 - />
211 - <TooltipPopup side="top">
212 - {label}
213 - {props.shortcutLabel ? ` (${props.shortcutLabel})` : ""}
214 - </TooltipPopup>
215 - </Tooltip>
103 + <ShortcutTooltip label={label} shortcut={props.shortcutLabel}>
104 + <Button
105 + type="button"
106 + size={props.size === "compact" ? "icon-xs" : "icon-sm"}
107 + variant="ghost"
108 + aria-label={label}
109 + disabled={props.disabled}
110 + onPointerDown={keepFocus}
111 + onClick={() => {
112 + if (!props.disabled) props.onStart();
113 + }}
114 + className="shrink-0"
115 + >
116 + <MicIcon />
117 + </Button>
118 + </ShortcutTooltip>
119 );
120 }
121
122 export function ComposerSpeechRecordingPill(props: {
123 state: VoiceInputState;
221 - progress: { downloaded: number; total: number } | null;
124 + progress: DownloadProgress;
125 finishShortcutLabel?: string | null;
126 cancelShortcutLabel?: string | null;
127 level: number;
128 recordingLimitSeconds?: number | undefined;
129 + /** Compact matches the `xs` buttons beside text-field editors. */
130 + size?: "default" | "compact";
131 onStart(): void;
132 disabled?: boolean;
133 onStop(): void;
134 onCancel(): void;
135 onSkipPostProcessing(): void;
136 }) {
232 - const presentation = resolveSpeechPresentation(props.state, props.progress);
233 - if (props.state.phase === "error") {
137 + const { phase } = props.state;
138 + const compact = props.size === "compact";
139 + if (phase === "error") {
140 return (
141 <VoiceInputErrorPill
142 + size={props.size}
143 error={props.state.error}
144 errorAction={props.state.errorAction}
145 disabled={props.disabled ?? false}
@@ -242,61 +149,74 @@ export function ComposerSpeechRecordingPill(props: {
149 />
150 );
151 }
245 - if (!presentation.showsConfirm) return null;
246 -
247 - const label = presentation.confirmEnabled
248 - ? "Finish voice input"
249 - : (presentation.status ?? "Voice input is busy");
152 + const presentation = resolveSpeechPresentation(props.state, props.progress);
153 + if (!presentation.active) return null;
154 + const recording = phase === "recording";
155 + const trailing = recording || phase === "post-processing";
156
157 return (
252 - <VoiceInputPill>
253 - <ComposerSpeechCancelButton
254 - state={props.state}
255 - shortcutLabel={props.cancelShortcutLabel ?? null}
256 - onCancel={props.onCancel}
257 - />
258 - <ComposerSpeechStatus
259 - state={props.state}
260 - progress={props.progress}
261 - level={props.level}
262 - recordingLimitSeconds={props.recordingLimitSeconds}
263 - />
264 - {props.state.phase === "post-processing" ? (
158 + <VoiceInputPill size={props.size}>
159 + <ShortcutTooltip label="Cancel voice input" shortcut={props.cancelShortcutLabel}>
160 <Button
161 type="button"
267 - size="sm-pill"
162 + size={compact ? "icon-xs" : "icon-sm-pill"}
163 variant="ghost"
269 - onPointerDown={(event) => event.preventDefault()}
164 + aria-label="Cancel voice input"
165 + onPointerDown={keepFocus}
166 + onClick={props.onCancel}
167 + className="shrink-0"
168 + >
169 + <XIcon />
170 + </Button>
171 + </ShortcutTooltip>
172 + <div
173 + className={cn(
174 + "flex min-w-0 flex-1 items-center gap-2 overflow-hidden",
175 + !trailing && "pe-2.5",
176 + )}
177 + role="status"
178 + aria-live={recording ? "off" : "polite"}
179 + aria-label={presentation.status ?? undefined}
180 + >
181 + {recording ? (
182 + <RecordingStatus
183 + level={props.level}
184 + recordingLimitSeconds={props.recordingLimitSeconds}
185 + />
186 + ) : (
187 + <>
188 + <Spinner aria-hidden size="sm" tone="muted" className="shrink-0" />
189 + <span className="min-w-0 truncate text-sm text-muted-foreground">
190 + {presentation.status}
191 + </span>
192 + </>
193 + )}
194 + </div>
195 + {phase === "post-processing" ? (
196 + <Button
197 + type="button"
198 + size={compact ? "xs" : "sm-pill"}
199 + variant="ghost"
200 + onPointerDown={keepFocus}
201 onClick={props.onSkipPostProcessing}
202 className="shrink-0"
203 >
204 Skip
205 </Button>
275 - ) : (
276 - <Tooltip>
277 - <TooltipTrigger
278 - render={
279 - <Button
280 - type="button"
281 - size="icon-sm-pill"
282 - aria-label={label}
283 - disabled={!presentation.confirmEnabled}
284 - onPointerDown={(event) => event.preventDefault()}
285 - onClick={props.onStop}
286 - className="shrink-0"
287 - >
288 - {presentation.confirmEnabled ? <CheckIcon /> : <Spinner aria-hidden />}
289 - </Button>
290 - }
291 - />
292 - <TooltipPopup side="top">
293 - {label}
294 - {presentation.confirmEnabled && props.finishShortcutLabel
295 - ? ` (${props.finishShortcutLabel})`
296 - : ""}
297 - </TooltipPopup>
298 - </Tooltip>
299 - )}
206 + ) : recording ? (
207 + <ShortcutTooltip label="Finish voice input" shortcut={props.finishShortcutLabel}>
208 + <Button
209 + type="button"
210 + size={compact ? "icon-xs" : "icon-sm-pill"}
211 + aria-label="Finish voice input"
212 + onPointerDown={keepFocus}
213 + onClick={props.onStop}
214 + className="shrink-0"
215 + >
216 + <CheckIcon />
217 + </Button>
218 + </ShortcutTooltip>
219 + ) : null}
220 </VoiceInputPill>
221 );
222 }
apps/web/src/components/chat/VoiceInputErrorPill.tsx
+55 -52
@@ -1,70 +1,74 @@
1 import { MICROPHONE_PERMISSION_ERROR } from "@t3tools/client-runtime/voice-input";
2 import { useNavigate } from "@tanstack/react-router";
3 -import { RotateCcwIcon, SettingsIcon, XIcon } from "lucide-react";
3 +import { CircleAlertIcon, XIcon } from "lucide-react";
4
5 import { Button } from "../ui/button";
6 import { Tooltip, TooltipPopup, TooltipTrigger } from "../ui/tooltip";
7 import { VoiceInputPill } from "./VoiceInputPill";
8
9 const VOICE_ERROR_LABELS: Record<string, string> = {
10 - [MICROPHONE_PERMISSION_ERROR]: "Microphone access needed",
11 - "Microphone access is required for voice input.": "Microphone access needed",
10 + [MICROPHONE_PERMISSION_ERROR]: "Mic blocked",
11 + "Microphone access is required for voice input.": "Mic blocked",
12 "This draft is no longer available.": "Draft unavailable",
13 - "Another voice recording is already active.": "Another recording active",
13 + "Another voice recording is already active.": "Already recording",
14 "Voice transcription is not available.": "Voice input unavailable",
15 - "Voice transcription is still finishing. Try again shortly.": "Transcription still busy",
15 + "Voice transcription is still finishing. Try again shortly.": "Still transcribing",
16 "Voice transcription is not available for this language.": "Language unsupported",
17 - "Could not prepare voice transcription.": "Couldn’t prepare dictation",
18 - "Could not start voice recording.": "Couldn’t start recording",
17 + "Could not prepare voice transcription.": "Couldn’t start",
18 + "Could not start voice recording.": "Couldn’t start",
19 "Voice recording was interrupted.": "Recording interrupted",
20 - "Could not finish voice recording.": "Couldn’t finish recording",
20 + "Could not finish voice recording.": "Couldn’t finish",
21 "Could not transcribe this recording.": "Transcription failed",
22 "Voice transcription on this environment failed.": "Transcription failed",
23 "The draft changed while voice input was running. The transcript was not added.": "Draft changed",
24 - "No speech was detected.": "No speech detected",
25 - "Microphone processing failed.": "Microphone processing failed",
26 - "The microphone disconnected.": "Microphone disconnected",
24 + "No speech was detected.": "No speech heard",
25 + "Microphone processing failed.": "Mic failed",
26 + "The microphone disconnected.": "Mic disconnected",
27 };
28
29 -function voiceErrorLabel(error: string | null): string {
30 - return (error && VOICE_ERROR_LABELS[error]) || "Voice input failed";
31 -}
29 +const RETRYABLE_TRANSCRIPTIONS = new Set([
30 + "Could not transcribe this recording.",
31 + "Voice transcription on this environment failed.",
32 +]);
33
34 export function VoiceInputErrorPill(props: {
35 error: string | null;
36 errorAction: "retry" | "settings" | null;
37 disabled?: boolean;
38 cancelShortcutLabel?: string | null;
39 + size?: "default" | "compact" | undefined;
40 onRetry(): void;
41 onDismiss(): void;
42 }) {
43 const navigate = useNavigate();
44 + const message = props.error ?? "Voice input failed";
45 const openSettings = props.errorAction === "settings";
43 - const actionLabel = openSettings
44 - ? "Open voice settings"
45 - : props.error === "Could not transcribe this recording." ||
46 - props.error === "Voice transcription on this environment failed."
47 - ? "Retry transcription"
48 - : "Record again";
46 + const compact = props.size === "compact";
47 + const action = openSettings
48 + ? { label: "Open voice settings", text: "Settings" }
49 + : props.error && RETRYABLE_TRANSCRIPTIONS.has(props.error)
50 + ? { label: "Retry transcription", text: "Retry" }
51 + : { label: "Record again", text: "Record again" };
52 return (
50 - <VoiceInputPill>
53 + <VoiceInputPill tone="error" size={props.size}>
54 <Tooltip>
55 <TooltipTrigger
56 render={
57 <Button
58 type="button"
56 - size="icon-sm-pill"
59 + size={compact ? "icon-xs" : "icon-sm-pill"}
60 variant="ghost"
61 aria-label="Dismiss voice input error"
62 onPointerDown={(event) => event.preventDefault()}
63 onClick={props.onDismiss}
64 + className="shrink-0"
65 >
66 <XIcon />
67 </Button>
68 }
69 />
70 <TooltipPopup side="top">
67 - Dismiss voice input error
71 + Dismiss
72 {props.cancelShortcutLabel ? ` (${props.cancelShortcutLabel})` : ""}
73 </TooltipPopup>
74 </Tooltip>
@@ -74,42 +78,41 @@ export function VoiceInputErrorPill(props: {
78 <span
79 tabIndex={0}
80 role="status"
77 - aria-label={props.error ?? "Voice input failed"}
78 - className="min-w-0 flex-1 truncate text-sm text-destructive"
81 + aria-label={message}
82 + className="flex min-w-0 items-center gap-1.5 rounded-sm text-sm font-medium outline-none focus-visible:ring-2 focus-visible:ring-ring focus-visible:ring-inset"
83 />
84 }
85 >
82 - {voiceErrorLabel(props.error)}
86 + <CircleAlertIcon aria-hidden className="size-3.5 shrink-0" />
87 + <span className="truncate">
88 + {(props.error && VOICE_ERROR_LABELS[props.error]) || "Voice input failed"}
89 + </span>
90 </TooltipTrigger>
84 - <TooltipPopup side="top">{props.error ?? "Voice input failed"}</TooltipPopup>
91 + <TooltipPopup side="top">{message}</TooltipPopup>
92 </Tooltip>
93 {props.errorAction ? (
87 - <Tooltip>
88 - <TooltipTrigger
89 - render={
90 - <Button
91 - type="button"
92 - size="icon-sm-pill"
93 - variant="ghost"
94 - aria-label={actionLabel}
95 - disabled={props.disabled}
96 - onPointerDown={(event) => event.preventDefault()}
97 - onClick={() => {
98 - if (openSettings) {
99 - props.onDismiss();
100 - void navigate({ to: "/settings/voice" });
101 - } else {
102 - props.onRetry();
103 - }
104 - }}
105 - >
106 - {openSettings ? <SettingsIcon /> : <RotateCcwIcon />}
107 - </Button>
94 + <Button
95 + type="button"
96 + size={compact ? "xs" : "sm-pill"}
97 + variant="outline"
98 + aria-label={action.label}
99 + disabled={props.disabled}
100 + onPointerDown={(event) => event.preventDefault()}
101 + onClick={() => {
102 + if (openSettings) {
103 + props.onDismiss();
104 + void navigate({ to: "/settings/voice" });
105 + } else {
106 + props.onRetry();
107 }
109 - />
110 - <TooltipPopup side="top">{actionLabel}</TooltipPopup>
111 - </Tooltip>
112 - ) : null}
108 + }}
109 + className="ms-auto shrink-0"
110 + >
111 + {action.text}
112 + </Button>
113 + ) : (
114 + <span aria-hidden className="w-1.5 shrink-0" />
115 + )}
116 </VoiceInputPill>
117 );
118 }
apps/web/src/components/chat/VoiceInputPill.tsx
+31 -9
@@ -1,8 +1,10 @@
1 import { memo, useEffect, useRef, type ReactNode } from "react";
2 import { useMediaQuery } from "~/hooks/useMediaQuery";
3 +import { cn } from "~/lib/utils";
4
4 -const WAVEFORM_BAR_COUNT = 28;
5 +const WAVEFORM_BAR_COUNT = 24;
6 const WAVEFORM_BAR_IDS = Array.from({ length: WAVEFORM_BAR_COUNT }, (_, index) => String(index));
7 +const WAVEFORM_REST_SCALE = 0.2;
8
9 export const VoiceWaveform = memo(function VoiceWaveform(props: { level: number }) {
10 const prefersReducedMotion = useMediaQuery("(prefers-reduced-motion: reduce)");
@@ -17,15 +19,15 @@ export const VoiceWaveform = memo(function VoiceWaveform(props: { level: number
19 barsRef.current.forEach((bar, index) => {
20 if (!bar) return;
21 const level = levels[index] ?? 0;
20 - bar.style.opacity = String(0.22 + level * 0.78);
21 - bar.style.transform = `scaleY(${prefersReducedMotion ? 0.35 : Math.max(0.08, level)})`;
22 + bar.style.opacity = String(0.3 + level * 0.55);
23 + bar.style.transform = `scaleY(${prefersReducedMotion ? 0.4 : Math.max(WAVEFORM_REST_SCALE, level)})`;
24 });
25 }, [prefersReducedMotion, props.level]);
26
27 return (
28 <div
29 aria-hidden
28 - className="flex h-5 min-w-0 flex-1 items-center justify-between gap-px overflow-hidden"
30 + className="flex h-4 min-w-0 flex-1 items-center justify-end gap-0.5 overflow-hidden"
31 >
32 {WAVEFORM_BAR_IDS.map((id, index) => (
33 <span
@@ -33,20 +35,40 @@ export const VoiceWaveform = memo(function VoiceWaveform(props: { level: number
35 ref={(bar) => {
36 barsRef.current[index] = bar;
37 }}
36 - className="h-full w-0.5 shrink-0 origin-center rounded-full bg-primary opacity-25 transition-[transform,opacity] duration-100 ease-out motion-reduce:transition-none"
37 - style={{ transform: `scaleY(${prefersReducedMotion ? 0.35 : 0.08})` }}
38 + className="h-full w-0.5 shrink-0 origin-center rounded-full bg-foreground opacity-35 transition-[transform,opacity] duration-100 ease-out motion-reduce:transition-none"
39 + style={{ transform: `scaleY(${prefersReducedMotion ? 0.4 : WAVEFORM_REST_SCALE})` }}
40 />
41 ))}
42 </div>
43 );
44 });
45
44 -export function VoiceInputPill({ children }: { children: ReactNode }) {
46 +/**
47 + * The capsule every dictation state renders in. Default matches the composer's send button
48 + * and insets its pill buttons; compact matches the `xs` buttons in text-field action rows.
49 + */
50 +export function VoiceInputPill({
51 + children,
52 + tone = "default",
53 + size = "default",
54 +}: {
55 + children: ReactNode;
56 + tone?: "default" | "error";
57 + size?: "default" | "compact" | undefined;
58 +}) {
59 return (
46 - // Same height as the toolbar's icon buttons, so starting a recording never resizes the composer.
60 <div
61 data-slot="voice-input-pill"
49 - className="flex h-8 w-48 min-w-0 shrink-0 items-center gap-2 sm:h-7 sm:w-64"
62 + className={cn(
63 + "flex min-w-0 items-center gap-1.5 ring-1 ring-inset",
64 + size === "compact"
65 + ? // Phones give the capsule its own row so the field's buttons stay together below it.
66 + "h-7 rounded-md max-sm:order-first max-sm:basis-full sm:h-6"
67 + : "h-9 max-w-fit grow-[1.5] basis-0 rounded-full p-0.5 sm:h-8",
68 + tone === "error"
69 + ? "bg-destructive/6 text-destructive ring-destructive/24"
70 + : "bg-muted/48 ring-border",
71 + )}
72 >
73 {children}
74 </div>
apps/web/src/components/chat/VoiceInputSetup.tsx
+151 -132
@@ -1,15 +1,25 @@
1 import { useStableCallback } from "~/hooks/useStableCallback";
2 -import { listBrowserMicrophones, type BrowserMicrophone } from "../../speech/browserMicrophones";
2 +import {
3 + listBrowserMicrophones,
4 + updateMicrophoneOrder,
5 + type BrowserMicrophone,
6 +} from "../../speech/browserMicrophones";
7 import type {
8 EnvironmentId,
9 EnvironmentSpeechModel,
10 EnvironmentSpeechStatus,
11 } from "@t3tools/contracts";
12 import { useNavigate } from "@tanstack/react-router";
9 -import { RefreshCwIcon } from "lucide-react";
13 +import { CircleAlertIcon, MicIcon, RefreshCwIcon } from "lucide-react";
14 import { useEffect, useState } from "react";
15
12 -import { useClientSettings, updateClientSettings } from "../../hooks/useSettings";
16 +import { useClientSettings } from "../../hooks/useSettings";
17 +import {
18 + isSystemDefaultDeviceId,
19 + microphoneOrder,
20 + preferMicrophone,
21 + preferredMicrophoneId,
22 +} from "../../speech/microphoneOrder";
23 import { useEnvironment } from "../../state/environments";
24 import { AuthSettingsWriteScope } from "@t3tools/contracts";
25 import { useEnvironmentScope } from "../../state/session";
@@ -17,7 +27,7 @@ import { languageLabel } from "../settings/SpeechLanguageSelect";
27 import { Button } from "../ui/button";
28 import { Dialog } from "../ui/dialog";
29 import { Select, SelectItem, SelectPopup, SelectTrigger, SelectValue } from "../ui/select";
20 -import { WizardFooter, WizardHeader, WizardPanel, WizardPopup, WizardSteps } from "../ui/wizard";
30 +import { WizardFooter, WizardHeader, WizardPanel, WizardPopup } from "../ui/wizard";
31
32 export function VoiceInputSetup(props: {
33 environmentId: EnvironmentId | null;
@@ -38,8 +48,9 @@ export function VoiceInputSetup(props: {
48 const navigate = useNavigate();
49 const environment = useEnvironment(props.environmentId);
50 const canManage = useEnvironmentScope(props.environmentId, AuthSettingsWriteScope);
41 - const selectedMicrophone = useClientSettings((settings) => settings.voiceMicrophone);
51 + const order = useClientSettings(microphoneOrder);
52 const [microphones, setMicrophones] = useState<BrowserMicrophone[]>([]);
53 + const selectedMicrophone = preferredMicrophoneId(order, microphones);
54 const [loadingMicrophones, setLoadingMicrophones] = useState(false);
55 const [microphoneError, setMicrophoneError] = useState<string | null>(null);
56 const refreshMicrophones = useStableCallback(async () => {
@@ -49,7 +60,7 @@ export function VoiceInputSetup(props: {
60 setMicrophones(await listBrowserMicrophones());
61 setMicrophoneError(null);
62 } catch {
52 - setMicrophoneError("Could not list microphones. You can still use the system default.");
63 + setMicrophoneError("Couldn’t list microphones. The system default still works.");
64 }
65 setLoadingMicrophones(false);
66 });
@@ -77,39 +88,38 @@ export function VoiceInputSetup(props: {
88 void navigate({ to: "/settings/voice", hash });
89 };
90
91 + const destination = environment?.label ?? "your transcription environment";
92 +
93 return (
94 <Dialog open={props.open} onOpenChange={props.onOpenChange}>
95 <WizardPopup>
96 <WizardHeader
84 - title="Set up voice input"
85 - description={`Transcribe speech on ${environment?.label ?? "your transcription environment"} and use voice input across environments.`}
86 - >
87 - <WizardSteps steps={["Model", "Ready to dictate"]} currentStep={props.step} />
88 - </WizardHeader>
97 + title={props.step === 0 ? "Set up voice input" : "Voice input is ready"}
98 + description={
99 + props.step === 0
100 + ? `Download a speech model once. Your audio is transcribed on ${destination}.`
101 + : "Pick a microphone, then start talking."
102 + }
103 + />
104 <WizardPanel>
105 {props.step === 0 ? (
106 <section className="space-y-3 text-sm">
107 {model ? (
93 - <div className="rounded-lg border border-border/70 bg-card/60 p-4">
94 - <div className="flex items-center justify-between gap-2">
95 - <h3 className="font-medium">{model.name}</h3>
96 - {model?.recommended && props.language ? (
97 - <span className="text-xs font-medium text-primary">
98 - Recommended for {languageLabel(props.language)}
99 - </span>
100 - ) : null}
108 + <div className="space-y-3 rounded-lg border border-border/70 bg-card/60 p-3">
109 + <div className="flex items-baseline justify-between gap-3">
110 + <h3 className="min-w-0 truncate font-medium">{model.name}</h3>
111 + <span className="shrink-0 text-xs text-muted-foreground tabular-nums">
112 + {Math.round(model.size / 1024 / 1024)} MB
113 + </span>
114 </div>
102 - <p className="mt-2 text-muted-foreground">
103 - Download the {Math.round(model.size / 1024 / 1024)} MB model to{" "}
104 - {environment?.label ?? "your transcription environment"}. Microphone audio is
105 - sent there for transcription, as chosen in Voice settings.
106 - </p>
107 - <p className="mt-2 text-muted-foreground">
108 - Silero (1.2 MB), used for silence filtering and pause detection, downloads
109 - alongside the speech model. An installed copy is reused.
115 + <p className="text-xs text-muted-foreground">
116 + {model.recommended && props.language
117 + ? `Recommended for ${languageLabel(props.language)}. `
118 + : ""}
119 + Includes a 1.2 MB silence detector.
120 </p>
121 {props.downloading ? (
112 - <div role="status" className="mt-4 space-y-1 text-muted-foreground">
122 + <div role="status" className="space-y-1.5">
123 <div
124 role="progressbar"
125 aria-label="Speech model download progress"
@@ -123,9 +133,9 @@ export function VoiceInputSetup(props: {
133 style={{ width: `${progress ?? 0}%` }}
134 />
135 </div>
126 - <p className="text-xs">
127 - {model?.state === "verifying"
128 - ? "Verifying download…"
136 + <p className="text-xs text-muted-foreground tabular-nums">
137 + {model.state === "verifying"
138 + ? "Verifying…"
139 : progress === null
140 ? "Starting download…"
141 : `${progress}% downloaded`}
@@ -139,119 +149,128 @@ export function VoiceInputSetup(props: {
149 </p>
150 ) : null}
151 {props.error ? (
142 - <p role="alert" className="text-destructive">
152 + <p role="alert" className="flex items-start gap-1.5 text-destructive">
153 + <CircleAlertIcon aria-hidden className="mt-0.5 size-3.5 shrink-0" />
154 {props.error}
155 </p>
156 ) : null}
157 </section>
158 ) : (
148 - <section className="space-y-4 text-sm">
149 - <div className="space-y-2">
150 - <label className="font-medium" htmlFor="voice-setup-microphone">
151 - Microphone
152 - </label>
153 - <div className="flex items-center gap-2">
154 - <Select
155 - value={selectedMicrophone ? `device:${selectedMicrophone}` : "system-default"}
156 - disabled={loadingMicrophones}
157 - onValueChange={(value) => {
158 - if (value)
159 - void updateClientSettings({
160 - voiceMicrophone:
161 - value === "system-default" ? "" : value.slice("device:".length),
162 - });
163 - }}
164 - >
165 - <SelectTrigger
166 - id="voice-setup-microphone"
167 - size="sm"
168 - aria-label="Microphone"
169 - className="min-w-0 flex-1"
170 - >
171 - <SelectValue>{selectedMicrophoneLabel}</SelectValue>
172 - </SelectTrigger>
173 - <SelectPopup align="end" alignItemWithTrigger={false}>
174 - <SelectItem value="system-default">System default</SelectItem>
175 - {selectedIsUnavailable ? (
176 - <SelectItem value={`device:${selectedMicrophone}`}>
177 - Selected microphone (Unavailable)
178 - </SelectItem>
179 - ) : null}
180 - {microphones.map((device) => (
181 - <SelectItem key={device.deviceId} value={`device:${device.deviceId}`}>
182 - {device.label}
183 - </SelectItem>
184 - ))}
185 - </SelectPopup>
186 - </Select>
187 - <Button
188 - type="button"
189 - variant="ghost"
190 - size="icon-sm"
191 - disabled={loadingMicrophones}
192 - aria-label="Refresh microphones"
193 - onClick={() => void refreshMicrophones()}
159 + <section className="space-y-2 text-sm">
160 + <label className="sr-only" htmlFor="voice-setup-microphone">
161 + Microphone
162 + </label>
163 + <div className="flex items-center gap-2">
164 + <Select
165 + value={selectedMicrophone ? `device:${selectedMicrophone}` : "system-default"}
166 + disabled={loadingMicrophones}
167 + onValueChange={(value) => {
168 + if (!value) return;
169 + const deviceId = value.slice("device:".length);
170 + const label =
171 + microphones.find((device) => device.deviceId === deviceId)?.label ?? "";
172 + updateMicrophoneOrder((order) =>
173 + preferMicrophone(
174 + order,
175 + value === "system-default" || isSystemDefaultDeviceId(deviceId)
176 + ? { kind: "default" }
177 + : { kind: "device", deviceId, label },
178 + ),
179 + );
180 + }}
181 + >
182 + <SelectTrigger
183 + id="voice-setup-microphone"
184 + size="sm"
185 + aria-label="Microphone"
186 + className="min-w-0 flex-1"
187 >
195 - <RefreshCwIcon className="size-3.5" />
196 - </Button>
197 - </div>
198 - {selectedIsUnavailable ? (
199 - <p className="text-xs text-destructive">
200 - The selected microphone is unavailable. Choose another or use the system
201 - default.
202 - </p>
203 - ) : null}
204 - {microphoneError ? (
205 - <p role="alert" className="text-xs text-muted-foreground">
206 - {microphoneError}
207 - </p>
208 - ) : null}
188 + <MicIcon aria-hidden className="size-3.5 shrink-0 text-muted-foreground" />
189 + <SelectValue>{selectedMicrophoneLabel}</SelectValue>
190 + </SelectTrigger>
191 + <SelectPopup align="end" alignItemWithTrigger={false}>
192 + <SelectItem value="system-default">System default</SelectItem>
193 + {selectedIsUnavailable ? (
194 + <SelectItem value={`device:${selectedMicrophone}`}>
195 + Selected microphone (Unavailable)
196 + </SelectItem>
197 + ) : null}
198 + {microphones.map((device) => (
199 + <SelectItem key={device.deviceId} value={`device:${device.deviceId}`}>
200 + {device.label}
201 + </SelectItem>
202 + ))}
203 + </SelectPopup>
204 + </Select>
205 + <Button
206 + type="button"
207 + variant="ghost"
208 + size="icon-sm"
209 + disabled={loadingMicrophones}
210 + aria-label="Refresh microphones"
211 + onClick={() => void refreshMicrophones()}
212 + >
213 + <RefreshCwIcon className="size-3.5" />
214 + </Button>
215 </div>
210 - <p className="text-muted-foreground">
211 - Speak and transcribe in multiple languages or translate into English, and get names
212 - and technical terms right with the dictionary in Voice settings.
213 - </p>
216 + {selectedIsUnavailable ? (
217 + <p className="text-xs text-destructive">
218 + That microphone is unplugged. Pick another one.
219 + </p>
220 + ) : null}
221 + {microphoneError ? (
222 + <p role="alert" className="text-xs text-muted-foreground">
223 + {microphoneError}
224 + </p>
225 + ) : null}
226 </section>
227 )}
228 </WizardPanel>
217 - <WizardFooter>
218 - {props.step === 0 ? (
219 - <>
220 - <Button variant="outline" onClick={() => openSettings("speech-models")}>
221 - {model ? "Choose another model or language" : "Choose a model or language"}
229 + {props.step === 0 ? (
230 + <WizardFooter
231 + leading={
232 + <Button variant="ghost" onClick={() => openSettings("speech-models")}>
233 + {model ? "Other models" : "Choose a model"}
234 </Button>
223 - {props.onRetryCatalog ? (
224 - <Button onClick={props.onRetryCatalog}>Retry loading models</Button>
225 - ) : null}
226 - {props.downloading ? (
227 - <Button variant="outline" disabled={!canManage} onClick={props.onCancelDownload}>
228 - Cancel download
229 - </Button>
230 - ) : null}
231 - {model ? (
232 - <Button
233 - disabled={!canManage || props.downloading || !status}
234 - onClick={props.onDownload}
235 - >
236 - {props.downloading
237 - ? "Downloading…"
238 - : props.error
239 - ? "Try again"
240 - : canManage
241 - ? "Download model"
242 - : "Ask an environment admin to download"}
243 - </Button>
244 - ) : null}
245 - </>
246 - ) : (
247 - <>
248 - <Button variant="outline" onClick={() => openSettings("")}>
249 - Customize in Voice settings
235 + }
236 + >
237 + {props.onRetryCatalog ? (
238 + <Button onClick={props.onRetryCatalog}>Retry loading models</Button>
239 + ) : null}
240 + {props.downloading ? (
241 + <Button variant="outline" disabled={!canManage} onClick={props.onCancelDownload}>
242 + Cancel download
243 </Button>
251 - <Button onClick={props.onStartRecording}>Start dictating</Button>
252 - </>
253 - )}
254 - </WizardFooter>
244 + ) : null}
245 + {model ? (
246 + <Button
247 + disabled={!canManage || props.downloading || !status}
248 + onClick={props.onDownload}
249 + >
250 + {props.downloading
251 + ? "Downloading…"
252 + : props.error
253 + ? "Try again"
254 + : canManage
255 + ? "Download model"
256 + : "Needs an admin to download"}
257 + </Button>
258 + ) : null}
259 + </WizardFooter>
260 + ) : (
261 + <WizardFooter
262 + leading={
263 + <Button variant="ghost" onClick={() => openSettings("")}>
264 + Voice settings
265 + </Button>
266 + }
267 + >
268 + <Button onClick={props.onStartRecording}>
269 + <MicIcon />
270 + Start dictating
271 + </Button>
272 + </WizardFooter>
273 + )}
274 </WizardPopup>
275 </Dialog>
276 );
apps/web/src/components/pullRequest/PullRequestCommentForm.tsx
+1 -3
@@ -137,9 +137,7 @@ export function PullRequestCommentForm({
137 }}
138 />
139 <div className="flex flex-wrap items-center justify-end gap-2">
140 - <div className="mr-auto flex min-w-0 flex-wrap items-center gap-1">
141 - <TextFieldSpeechControls speech={speech} disabled={locked} />
142 - </div>
140 + <TextFieldSpeechControls speech={speech} disabled={locked} />
141 {followUpAction === null ? null : (
142 <Button
143 size="xs"
apps/web/src/components/pullRequest/PullRequestReviewForm.tsx
+1 -3
@@ -177,9 +177,7 @@ export function PullRequestReviewForm({
177 }}
178 />
179 <div className="mt-2 flex flex-wrap items-center justify-end gap-2">
180 - <div className="mr-auto flex min-w-0 flex-wrap items-center gap-1">
181 - <TextFieldSpeechControls speech={speech} disabled={pending} />
182 - </div>
180 + <TextFieldSpeechControls speech={speech} disabled={pending} />
181 {offered.length > 1 ? (
182 <Select
183 value={selectedVerdict?.value ?? null}
apps/web/src/components/settings/MicrophoneOrderList.tsx new
+172
@@ -0,0 +1,172 @@
1 +import {
2 + DndContext,
3 + type DragEndEvent,
4 + PointerSensor,
5 + closestCenter,
6 + useSensor,
7 + useSensors,
8 +} from "@dnd-kit/core";
9 +import { restrictToParentElement, restrictToVerticalAxis } from "@dnd-kit/modifiers";
10 +import { SortableContext, useSortable, verticalListSortingStrategy } from "@dnd-kit/sortable";
11 +import { CSS } from "@dnd-kit/utilities";
12 +import { ArrowDownIcon, ArrowUpIcon, GripVerticalIcon, XIcon } from "lucide-react";
13 +
14 +import { cn } from "~/lib/utils";
15 +import { type BrowserMicrophone, updateMicrophoneOrder } from "../../speech/browserMicrophones";
16 +import {
17 + type MicrophoneEntry,
18 + forgetMicrophone,
19 + microphoneEntryKey,
20 + moveMicrophone,
21 + preferredMicrophone,
22 +} from "../../speech/microphoneOrder";
23 +import { Button } from "../ui/button";
24 +
25 +/** The saved microphone order. Keyboard users reorder with the step buttons. */
26 +export function MicrophoneOrderList(props: {
27 + readonly order: ReadonlyArray<MicrophoneEntry>;
28 + readonly devices: ReadonlyArray<BrowserMicrophone>;
29 +}) {
30 + const { order, devices } = props;
31 + const keys = order.map(microphoneEntryKey);
32 + const inUse = microphoneEntryKey(preferredMicrophone(order, devices));
33 + // Device ids stay hidden until microphone permission, so connection is unknown before it.
34 + const devicesKnown = devices.some((device) => device.deviceId);
35 + const sensors = useSensors(useSensor(PointerSensor, { activationConstraint: { distance: 4 } }));
36 + const move = (key: string, steps: number) =>
37 + updateMicrophoneOrder((current) => moveMicrophone(current, key, steps));
38 + const handleDragEnd = (event: DragEndEvent) => {
39 + if (event.over === null || event.active.id === event.over.id) return;
40 + const from = keys.indexOf(String(event.active.id));
41 + const to = keys.indexOf(String(event.over.id));
42 + if (from !== -1 && to !== -1) move(String(event.active.id), to - from);
43 + };
44 +
45 + return (
46 + <DndContext
47 + sensors={sensors}
48 + collisionDetection={closestCenter}
49 + modifiers={[restrictToVerticalAxis, restrictToParentElement]}
50 + onDragEnd={handleDragEnd}
51 + >
52 + <SortableContext items={keys} strategy={verticalListSortingStrategy}>
53 + <ol
54 + aria-label="Microphones, preferred first"
55 + className="divide-y divide-border/60 rounded-lg border border-border/60"
56 + >
57 + {order.map((entry, index) => {
58 + const key = microphoneEntryKey(entry);
59 + const device =
60 + entry.kind === "device"
61 + ? devices.find((candidate) => candidate.deviceId === entry.deviceId)
62 + : undefined;
63 + return (
64 + <MicrophoneRow
65 + key={key}
66 + id={key}
67 + name={
68 + entry.kind === "default"
69 + ? "System default"
70 + : entry.label || device?.label || "Saved microphone"
71 + }
72 + position={index + 1}
73 + count={order.length}
74 + inUse={key === inUse}
75 + notConnected={entry.kind === "device" && devicesKnown && !device}
76 + onMove={(steps) => move(key, steps)}
77 + onForget={
78 + entry.kind === "device"
79 + ? () => updateMicrophoneOrder((current) => forgetMicrophone(current, key))
80 + : undefined
81 + }
82 + />
83 + );
84 + })}
85 + </ol>
86 + </SortableContext>
87 + </DndContext>
88 + );
89 +}
90 +
91 +function MicrophoneRow(props: {
92 + readonly id: string;
93 + readonly name: string;
94 + readonly position: number;
95 + readonly count: number;
96 + readonly inUse: boolean;
97 + readonly notConnected: boolean;
98 + readonly onMove: (steps: number) => void;
99 + readonly onForget: (() => void) | undefined;
100 +}) {
101 + const { name } = props;
102 + const { listeners, setNodeRef, setActivatorNodeRef, transform, transition, isDragging } =
103 + useSortable({ id: props.id });
104 + return (
105 + <li
106 + ref={setNodeRef}
107 + style={{ transform: CSS.Translate.toString(transform), transition }}
108 + className={cn(
109 + // Phones put the marker and controls on a second line so names stay readable.
110 + "grid grid-cols-[1rem_minmax(0,1fr)_auto] items-center gap-x-2 py-1.5 ps-3 pe-1.5 @min-[32rem]/settings-row:grid-cols-[1rem_auto_minmax(0,1fr)_auto_auto]",
111 + isDragging && "relative z-10 rounded-lg bg-background shadow-md",
112 + )}
113 + >
114 + <span className="text-xs tabular-nums text-muted-foreground">{props.position}</span>
115 + <span className="col-span-2 text-sm leading-5 break-words @min-[32rem]/settings-row:col-span-1">
116 + {name}
117 + </span>
118 + <span className="col-start-2 flex @min-[32rem]/settings-row:col-start-auto">
119 + {props.inUse ? (
120 + <span className="rounded-sm bg-success/12 px-1 text-2xs text-success-foreground">
121 + In use
122 + </span>
123 + ) : props.notConnected ? (
124 + <span className="text-2xs text-muted-foreground">Not connected</span>
125 + ) : null}
126 + </span>
127 + <span
128 + ref={setActivatorNodeRef}
129 + {...listeners}
130 + aria-hidden
131 + className="hidden size-6 shrink-0 cursor-grab touch-none items-center justify-center rounded text-muted-foreground/70 hover:text-foreground active:cursor-grabbing @min-[32rem]/settings-row:flex"
132 + >
133 + <GripVerticalIcon className="size-3.5" />
134 + </span>
135 + <span className="flex items-center gap-0.5">
136 + <Button
137 + type="button"
138 + size="icon-xs"
139 + variant="ghost-muted"
140 + disabled={props.position === 1}
141 + aria-label={`Move ${name} up`}
142 + onClick={() => props.onMove(-1)}
143 + >
144 + <ArrowUpIcon />
145 + </Button>
146 + <Button
147 + type="button"
148 + size="icon-xs"
149 + variant="ghost-muted"
150 + disabled={props.position === props.count}
151 + aria-label={`Move ${name} down`}
152 + onClick={() => props.onMove(1)}
153 + >
154 + <ArrowDownIcon />
155 + </Button>
156 + {props.onForget ? (
157 + <Button
158 + type="button"
159 + size="icon-xs"
160 + variant="ghost-muted"
161 + aria-label={`Forget ${name}`}
162 + onClick={props.onForget}
163 + >
164 + <XIcon />
165 + </Button>
166 + ) : (
167 + <span aria-hidden className="size-7 sm:size-6" />
168 + )}
169 + </span>
170 + </li>
171 + );
172 +}
apps/web/src/components/settings/MicrophoneTest.tsx
+113 -116
@@ -1,12 +1,15 @@
1 import { MICROPHONE_PERMISSION_ERROR } from "@t3tools/client-runtime/voice-input";
2 -import { CheckIcon, MicIcon, PauseIcon, PlayIcon, RotateCcwIcon, XIcon } from "lucide-react";
2 +import { MicIcon, PauseIcon, PlayIcon, RotateCcwIcon, SquareIcon, XIcon } from "lucide-react";
3 import { useEffect, useRef, useState } from "react";
4
5 import { useStableCallback } from "~/hooks/useStableCallback";
6 +import { openMicrophone } from "../../speech/browserPcmCapture";
7 import workletUrl from "../../speech/pcmWorklet.ts?worker&url";
7 -import { VoiceInputErrorPill } from "../chat/VoiceInputErrorPill";
8 -import { VoiceInputPill, VoiceWaveform } from "../chat/VoiceInputPill";
8 +import { VoiceWaveform } from "../chat/VoiceInputPill";
9 +import { cn } from "../../lib/utils";
10 import { Button } from "../ui/button";
11 +import { SettingsRow } from "./settingsLayout";
12 +import { searchableSetting } from "./settingsSearch";
13
14 const LOOPBACK_SECONDS = 10;
15 const formatTime = (seconds: number) =>
@@ -27,7 +30,14 @@ type Capture = {
30 * device: no environment, transcriber, or provider is involved, so a silent result points
31 * at the microphone rather than at transcription.
32 */
30 -export function MicrophoneTest({ microphoneId }: { microphoneId: string }) {
33 +export function MicrophoneTest({
34 + microphoneId,
35 + onMicrophoneAccess,
36 +}: {
37 + microphoneId: string;
38 + /** Labels and device ids become readable once the browser grants access. */
39 + onMicrophoneAccess: () => void;
40 +}) {
41 const [phase, setPhase] = useState<"idle" | "starting" | "recording" | "ready">("idle");
42 const [recording, setRecording] = useState<{ url: string; duration: number } | null>(null);
43 const [error, setError] = useState<string | null>(null);
@@ -82,14 +92,13 @@ export function MicrophoneTest({ microphoneId }: { microphoneId: string }) {
92 const current: Capture = {};
93 capture.current = current;
94 try {
85 - const stream = await navigator.mediaDevices.getUserMedia({
86 - audio: microphoneId ? { deviceId: { exact: microphoneId } } : true,
87 - });
95 + const stream = await openMicrophone(microphoneId);
96 if (capture.current !== current) {
97 stream.getTracks().forEach((track) => track.stop());
98 return;
99 }
100 current.stream = stream;
101 + onMicrophoneAccess();
102 const recorder = new MediaRecorder(stream);
103 current.recorder = recorder;
104 const chunks: Blob[] = [];
@@ -163,15 +172,15 @@ export function MicrophoneTest({ microphoneId }: { microphoneId: string }) {
172 if (capture.current !== current) return;
173 releaseCapture();
174 setPhase("idle");
175 + const name = typeof cause === "object" && cause !== null && "name" in cause && cause.name;
176 setError(
167 - typeof cause === "object" &&
168 - cause !== null &&
169 - "name" in cause &&
170 - cause.name === "NotAllowedError"
177 + name === "NotAllowedError"
178 ? MICROPHONE_PERMISSION_ERROR
172 - : cause instanceof Error
173 - ? cause.message
174 - : "Could not access the microphone.",
179 + : name === "NotFoundError"
180 + ? "No microphone found."
181 + : cause instanceof Error
182 + ? cause.message
183 + : "Could not access the microphone.",
184 );
185 }
186 };
@@ -181,70 +190,11 @@ export function MicrophoneTest({ microphoneId }: { microphoneId: string }) {
190 setError(null);
191 };
192
184 - if (error) {
185 - return (
186 - <VoiceInputErrorPill
187 - error={error}
188 - errorAction="retry"
189 - onRetry={() => void start()}
190 - onDismiss={cancel}
191 - />
192 - );
193 - }
194 - if (phase === "idle") {
195 - return (
196 - <Button size="sm" variant="outline" onClick={() => void start()}>
197 - <MicIcon />
198 - Test microphone
199 - </Button>
200 - );
201 - }
202 - if (phase === "ready" && recording) {
203 - return (
204 - <MicrophonePlayback
205 - key={recording.url}
206 - src={recording.url}
207 - duration={recording.duration}
208 - onRetry={() => void start()}
209 - onDone={cancel}
210 - />
211 - );
212 - }
213 - return (
214 - <VoiceInputPill>
193 + const control =
194 + phase === "recording" || phase === "starting" ? (
195 <Button
216 - size="icon-sm-pill"
217 - variant="ghost"
218 - aria-label="Cancel microphone test"
219 - onClick={cancel}
220 - >
221 - <XIcon />
222 - </Button>
223 - {phase === "recording" ? (
224 - <>
225 - <VoiceWaveform level={level} />
226 - <span aria-hidden className="size-1.5 shrink-0 rounded-full bg-primary" />
227 - <span
228 - className="shrink-0 text-xs tabular-nums text-muted-foreground"
229 - aria-label="Recording time"
230 - >
231 - {formatTime(elapsed)} / {formatTime(LOOPBACK_SECONDS)}
232 - </span>
233 - <meter
234 - aria-label="Microphone input level"
235 - min={0}
236 - max={1}
237 - value={level}
238 - className="sr-only"
239 - />
240 - </>
241 - ) : (
242 - <span role="status" className="min-w-0 flex-1 truncate text-xs text-muted-foreground">
243 - Opening microphone…
244 - </span>
245 - )}
246 - <Button
247 - size="icon-sm-pill"
196 + size="sm"
197 + variant="outline"
198 aria-label="Stop recording"
199 disabled={phase !== "recording"}
200 onClick={() => {
@@ -252,23 +202,77 @@ export function MicrophoneTest({ microphoneId }: { microphoneId: string }) {
202 if (recorder && recorder.state !== "inactive") recorder.stop();
203 }}
204 >
255 - <CheckIcon />
205 + <SquareIcon />
206 + Stop
207 + </Button>
208 + ) : (
209 + <Button size="sm" variant="outline" onClick={() => void start()}>
210 + {phase === "ready" ? <RotateCcwIcon /> : <MicIcon />}
211 + {phase === "ready" ? "Record again" : "Record"}
212 </Button>
257 - </VoiceInputPill>
213 + );
214 + const panel = error ? (
215 + <p role="alert" className="min-w-0 flex-1 text-xs text-destructive">
216 + {error}
217 + </p>
218 + ) : phase === "starting" ? (
219 + <span role="status" className="min-w-0 flex-1 truncate text-xs text-muted-foreground">
220 + Opening microphone…
221 + </span>
222 + ) : phase === "recording" ? (
223 + <>
224 + <span aria-hidden className="size-1.5 shrink-0 rounded-full bg-destructive" />
225 + <div className="flex w-24 min-w-0">
226 + <VoiceWaveform level={level} />
227 + </div>
228 + <span
229 + className="ms-auto shrink-0 text-xs tabular-nums text-muted-foreground"
230 + aria-label="Recording time"
231 + >
232 + {formatTime(elapsed)} / {formatTime(LOOPBACK_SECONDS)}
233 + </span>
234 + <meter
235 + aria-label="Microphone input level"
236 + min={0}
237 + max={1}
238 + value={level}
239 + className="sr-only"
240 + />
241 + </>
242 + ) : phase === "ready" && recording ? (
243 + <MicrophonePlayback key={recording.url} src={recording.url} duration={recording.duration} />
244 + ) : null;
245 +
246 + return (
247 + <SettingsRow
248 + {...searchableSetting("microphone-test")}
249 + title="Test microphone"
250 + description="Record ten seconds and play it back. The audio stays on this device."
251 + control={control}
252 + >
253 + {panel ? (
254 + <div
255 + className={cn(
256 + "mt-3 mb-2 flex min-h-10 items-center gap-3 rounded-lg border border-border/60 bg-muted/30 py-1 pe-1",
257 + phase === "ready" && !error ? "ps-1" : "ps-3",
258 + )}
259 + >
260 + {panel}
261 + <Button
262 + size="icon-sm"
263 + variant="ghost-muted"
264 + aria-label={phase === "ready" ? "Discard recording" : "Cancel microphone test"}
265 + onClick={cancel}
266 + >
267 + <XIcon />
268 + </Button>
269 + </div>
270 + ) : null}
271 + </SettingsRow>
272 );
273 }
274
261 -function MicrophonePlayback({
262 - src,
263 - duration,
264 - onRetry,
265 - onDone,
266 -}: {
267 - src: string;
268 - duration: number;
269 - onRetry(): void;
270 - onDone(): void;
271 -}) {
275 +function MicrophonePlayback({ src, duration }: { src: string; duration: number }) {
276 const audio = useRef<HTMLAudioElement>(null);
277 const [playing, setPlaying] = useState(false);
278 const [position, setPosition] = useState(0);
@@ -292,7 +296,7 @@ function MicrophonePlayback({
296 };
297
298 return (
295 - <div className="flex flex-col gap-1">
299 + <>
300 <audio
301 ref={audio}
302 src={src}
@@ -303,15 +307,19 @@ function MicrophonePlayback({
307 onTimeUpdate={() => setPosition(audio.current?.currentTime ?? 0)}
308 onError={() => setFailed(true)}
309 />
306 - <VoiceInputPill>
307 - <Button
308 - size="icon-sm-pill"
309 - variant="ghost"
310 - aria-label={playing ? "Pause recording" : "Play recording"}
311 - onClick={() => void togglePlayback()}
312 - >
313 - {playing ? <PauseIcon /> : <PlayIcon />}
314 - </Button>
310 + <Button
311 + size="icon-sm-pill"
312 + variant="outline"
313 + aria-label={playing ? "Pause recording" : "Play recording"}
314 + onClick={() => void togglePlayback()}
315 + >
316 + {playing ? <PauseIcon /> : <PlayIcon />}
317 + </Button>
318 + {failed ? (
319 + <p role="alert" className="min-w-0 flex-1 text-xs text-destructive">
320 + Could not play this recording. Try recording again.
321 + </p>
322 + ) : (
323 <input
324 type="range"
325 aria-label="Recording playback position"
@@ -330,21 +338,10 @@ function MicrophonePlayback({
338 }}
339 className="h-1 min-w-0 flex-1 cursor-pointer appearance-none rounded-full bg-muted accent-primary focus-visible:outline-2 focus-visible:-outline-offset-2 focus-visible:outline-ring [&::-webkit-slider-thumb]:size-2.5 [&::-webkit-slider-thumb]:appearance-none [&::-webkit-slider-thumb]:rounded-full [&::-webkit-slider-thumb]:bg-primary [&::-moz-range-thumb]:size-2.5 [&::-moz-range-thumb]:rounded-full [&::-moz-range-thumb]:border-0 [&::-moz-range-thumb]:bg-primary"
340 />
333 - <span className="shrink-0 text-xs tabular-nums text-muted-foreground">
334 - {formatTime(position)}
335 - </span>
336 - <Button size="icon-sm-pill" variant="ghost" aria-label="Record again" onClick={onRetry}>
337 - <RotateCcwIcon />
338 - </Button>
339 - <Button size="icon-sm-pill" aria-label="Done" onClick={onDone}>
340 - <CheckIcon />
341 - </Button>
342 - </VoiceInputPill>
343 - {failed ? (
344 - <p role="alert" className="text-xs text-destructive">
345 - Could not play this recording. Try recording again.
346 - </p>
347 - ) : null}
348 - </div>
341 + )}
342 + <span className="shrink-0 text-xs tabular-nums text-muted-foreground">
343 + {formatTime(position)} / {formatTime(duration)}
344 + </span>
345 + </>
346 );
347 }
apps/web/src/components/settings/SpeechModelComparison.tsx
-27
@@ -1,7 +1,5 @@
1 import type { EnvironmentSpeechModel, SpeechBenchmarkTier } from "@t3tools/contracts";
2 -import { CheckIcon } from "lucide-react";
2
4 -import { cn } from "../../lib/utils";
3 import { languageLabel } from "./SpeechLanguageSelect";
4
5 const commonLanguages = [
@@ -45,31 +43,6 @@ export function orderSpeechLanguages(languages: readonly string[], preferred: re
43 );
44 }
45
48 -export function LanguageRow({
49 - label,
50 - selected,
51 - onClick,
52 -}: {
53 - label: string;
54 - selected: boolean;
55 - onClick: () => void;
56 -}) {
57 - return (
58 - <button
59 - type="button"
60 - aria-pressed={selected}
61 - onClick={onClick}
62 - className={cn(
63 - "flex shrink-0 cursor-pointer items-center justify-between gap-2 rounded-md px-2.5 py-1.5 text-left text-sm outline-none hover:bg-muted/50 focus-visible:ring-2 focus-visible:ring-inset focus-visible:ring-ring",
64 - selected && "bg-muted font-medium",
65 - )}
66 - >
67 - {label}
68 - {selected && <CheckIcon className="size-3.5 shrink-0" />}
69 - </button>
70 - );
71 -}
72 -
46 export function ModelBenchmarks({
47 model,
48 languages,
apps/web/src/components/settings/VoiceSettingsPanel.browser.test.tsx
+56 -2
@@ -14,6 +14,7 @@ import { expect, it, vi } from "vite-plus/test";
14 import { userEvent } from "vite-plus/test/browser";
15 import { render } from "../../../test/render";
16 import { __setClientSettingsForTests, getClientSettings } from "../../hooks/useSettings";
17 +import { chooseRecordingMicrophone } from "../../speech/browserMicrophones";
18 import {
19 useEnvironmentSpeechInput,
20 useVoiceTranscriptionDestination,
@@ -331,17 +332,23 @@ it("tests the selected model after download and preserves its transcript across
332 const result = screen.getByRole("status", { name: "Test transcript" });
333 await expect.element(result).toHaveTextContent(transcript);
334
334 - const search = screen.getByRole("textbox", { name: "Search transcription models" });
335 + const search = screen.getByRole("searchbox", { name: "Search transcription models" });
336 await userEvent.fill(search, "no matching model");
337 await expect.element(screen.getByText("No models match.")).toBeVisible();
338 await userEvent.clear(search);
339 + const languages = screen.getByRole("button", { name: /^Filter by language/ });
340 + await userEvent.click(languages);
341 await userEvent.fill(screen.getByRole("textbox", { name: "Search languages" }), "fr");
339 - const french = screen.getByRole("button", { name: "French", exact: true });
342 + const french = screen.getByRole("menuitemcheckbox", { name: "French", exact: true });
343 await userEvent.click(french);
344 + await userEvent.keyboard("{Escape}");
345 await expect.element(screen.getByText("Bilingual", { exact: true })).toBeVisible();
346 await expect.element(screen.getByText("B+", { exact: true })).toBeVisible();
347 expect(screen.container.querySelector("#transcription-test")).toHaveAttribute("hidden");
348 + await userEvent.click(languages);
349 + await userEvent.fill(screen.getByRole("textbox", { name: "Search languages" }), "fr");
350 await userEvent.click(french);
351 + await userEvent.keyboard("{Escape}");
352 await expect.element(result).toBeVisible();
353 await vi.waitFor(() =>
354 expect(
@@ -591,3 +598,50 @@ it("keeps punctuation in a saved dictionary term and its alias", async () => {
598 { term: "O'Reilly", aliases: ['"Oh Really"'] },
599 ]);
600 });
601 +
602 +it("records from the first connected microphone in the saved order", async () => {
603 + const previous = getClientSettings();
604 + using cleanup = new DisposableStack();
605 + cleanup.defer(() => __setClientSettingsForTests(previous));
606 + // A single saved microphone from before the order existed comes first.
607 + __setClientSettingsForTests({ ...previous, voiceMicrophone: "yeti", voiceMicrophones: [] });
608 + const input = (deviceId: string, label: string) =>
609 + ({ deviceId, label, groupId: deviceId, kind: "audioinput" }) as MediaDeviceInfo;
610 + const yeti = input("yeti", "Blue Yeti");
611 + const airpods = input("airpods", "AirPods Pro");
612 + using devices = vi
613 + .spyOn(navigator.mediaDevices, "enumerateDevices")
614 + .mockResolvedValue([yeti, airpods]);
615 + const saved = () =>
616 + getClientSettings().voiceMicrophones.map((entry) =>
617 + entry.kind === "default" ? "System default" : entry.label,
618 + );
619 + await using screen = await render(
620 + <SettingsScopeProvider search={{}} onChange={() => {}}>
621 + <VoiceSettingsPanel />
622 + </SettingsScopeProvider>,
623 + );
624 + const list = screen.getByRole("list", { name: "Microphones, preferred first" });
625 + await expect.poll(saved).toEqual(["Blue Yeti", "System default", "AirPods Pro"]);
626 + const row = (index: number) => list.getByRole("listitem").nth(index);
627 + await expect.element(row(0)).toHaveTextContent("1Blue YetiIn use");
628 + expect(await chooseRecordingMicrophone()).toBe("yeti");
629 +
630 + devices.mockResolvedValue([airpods]);
631 + await userEvent.click(screen.getByRole("button", { name: "Refresh microphones" }));
632 + await expect.element(row(0)).toHaveTextContent("1Blue YetiNot connected");
633 + await expect.element(row(1)).toHaveTextContent("2System defaultIn use");
634 + expect(await chooseRecordingMicrophone()).toBe("");
635 +
636 + await userEvent.click(screen.getByRole("button", { name: "Move AirPods Pro up" }));
637 + await userEvent.click(screen.getByRole("button", { name: "Move AirPods Pro up" }));
638 + expect(saved()).toEqual(["AirPods Pro", "Blue Yeti", "System default"]);
639 + await expect.element(screen.getByRole("button", { name: "Move AirPods Pro up" })).toBeDisabled();
640 + expect(await chooseRecordingMicrophone()).toBe("airpods");
641 +
642 + await userEvent.click(screen.getByRole("button", { name: "Forget Blue Yeti" }));
643 + expect(saved()).toEqual(["AirPods Pro", "System default"]);
644 + await expect
645 + .element(screen.getByRole("button", { name: "Forget System default" }))
646 + .not.toBeInTheDocument();
647 +});
apps/web/src/components/settings/VoiceSettingsPanel.tsx
+514 -505
@@ -32,6 +32,7 @@ import { createModelSelection } from "@t3tools/shared/model";
32 import { resolveSpeechPostProcessingModelSelection } from "@t3tools/shared/serverSettings";
33 import { cloudSpeechLanguages } from "@t3tools/shared/speech";
34 import {
35 + ArrowDownUpIcon,
36 AudioLinesIcon,
37 CheckIcon,
38 DownloadIcon,
@@ -40,9 +41,10 @@ import {
41 MonitorIcon,
42 MicOffIcon,
43 MicIcon,
44 + PlusIcon,
45 RefreshCwIcon,
46 + SearchIcon,
47 StarIcon,
45 - SlidersHorizontalIcon,
48 Trash2Icon,
49 UnlinkIcon,
50 XIcon,
@@ -50,7 +52,17 @@ import {
52 import { useEffect, useRef, useState, type ReactNode } from "react";
53
54 import { useStableCallback } from "~/hooks/useStableCallback";
53 -import { listBrowserMicrophones, type BrowserMicrophone } from "../../speech/browserMicrophones";
55 +import { cn } from "~/lib/utils";
56 +import {
57 + listAudioInputs,
58 + updateMicrophoneOrder,
59 + type BrowserMicrophone,
60 +} from "../../speech/browserMicrophones";
61 +import {
62 + microphoneOrder,
63 + preferredMicrophoneId,
64 + rememberMicrophones,
65 +} from "../../speech/microphoneOrder";
66 import { preferredSpeechLanguages } from "../../speech/preferredSpeechLanguages";
67 import {
68 chooseAudioDestination,
@@ -91,6 +103,7 @@ import {
103 MenuRadioGroup,
104 MenuRadioItem,
105 MenuRadioItemIndicator,
106 + MenuSearchInput,
107 MenuSeparator,
108 MenuTrigger,
109 } from "../ui/menu";
@@ -106,9 +119,10 @@ import {
119 import { Switch } from "../ui/switch";
120 import { Textarea } from "../ui/textarea";
121 import { toastManager } from "../ui/toast";
122 +import { MicrophoneOrderList } from "./MicrophoneOrderList";
123 import { MicrophoneTest } from "./MicrophoneTest";
124 import { CloudVoiceConfiguration } from "./CloudVoiceConfiguration";
111 -import { LanguageRow, ModelBenchmarks, orderSpeechLanguages } from "./SpeechModelComparison";
125 +import { ModelBenchmarks, orderSpeechLanguages } from "./SpeechModelComparison";
126 import { languageLabel, SpeechLanguageSelect } from "./SpeechLanguageSelect";
127 import { filterSpeechModels, type SpeechModelSort } from "./speechModelBrowser";
128 import { AdvancedSettings } from "./AdvancedSettings";
@@ -136,12 +150,11 @@ function CloudSpeechLogo({ provider }: { provider: CloudSpeechProvider }) {
150 />
151 );
152 }
139 -const SYSTEM_DEFAULT = "system-default";
153 const formatSize = (bytes: number) => `${Math.round(bytes / 1024 / 1024)} MB`;
141 -const SHORTCUT_MODE_LABELS: Record<VoiceShortcutMode, string> = {
142 - auto: "Auto: press to toggle, hold to record",
143 - hold: "Hold: record while held",
144 - toggle: "Toggle: press to start, press again to stop",
154 +const SHORTCUT_MODES: Record<VoiceShortcutMode, { label: string; hint: string }> = {
155 + auto: { label: "Auto", hint: "Tap to toggle, hold to talk" },
156 + hold: { label: "Hold", hint: "Record while held" },
157 + toggle: { label: "Toggle", hint: "Tap to start, tap to stop" },
158 };
159 const UNLOAD_LABELS: Record<SpeechModelUnloadTimeout, string> = {
160 never: "Never",
@@ -171,6 +184,8 @@ type ModelCardProps = {
184 readonly onCancel: () => void;
185 readonly onDelete: () => void;
186 readonly testControls?: ReactNode;
187 + /** Keeps the controls under the text, so a live test never squeezes the model's details. */
188 + readonly stacked?: boolean;
189 readonly children?: ReactNode;
190 };
191
@@ -191,120 +206,128 @@ function ModelCard(props: ModelCardProps) {
206 const downloading = model.state === "downloading" || model.state === "verifying";
207 const progress = model.downloaded === undefined ? 0 : (model.downloaded / model.size) * 100;
208 return (
194 - <div className="flex flex-wrap items-start gap-3 px-3 py-2.5 sm:px-4">
195 - <div className="min-w-0 flex-1">
196 - <div className="flex flex-wrap items-center gap-1.5">
197 - <span className="text-sm font-medium leading-5">{model.name}</span>
198 - {model.recommended ? (
199 - <ModelBadge
200 - icon={<StarIcon className="mr-1 size-3" />}
201 - label="Recommended"
202 - tooltip="A good default pick for the languages it covers."
203 - />
209 + <div className="@container/model px-3 py-3">
210 + <div
211 + className={cn(
212 + "flex flex-col gap-2.5",
213 + !props.stacked &&
214 + "@min-[28rem]/model:flex-row @min-[28rem]/model:items-start @min-[28rem]/model:gap-4",
215 + )}
216 + >
217 + <div className="min-w-0 flex-1">
218 + <div className="flex flex-wrap items-center gap-1.5">
219 + <span className="text-sm font-medium leading-5">{model.name}</span>
220 + {model.recommended ? (
221 + <ModelBadge
222 + icon={<StarIcon className="mr-1 size-3" />}
223 + label="Recommended"
224 + tooltip="A good default pick for the languages it covers."
225 + />
226 + ) : null}
227 + {model.supportsStreaming ? (
228 + <ModelBadge
229 + icon={<AudioLinesIcon className="mr-1 size-3" />}
230 + label="Live preview"
231 + tooltip="Shows words in your draft as you speak, before the recording finishes."
232 + />
233 + ) : null}
234 + {model.supportsTranslation ? (
235 + <ModelBadge
236 + icon={<LanguagesIcon className="mr-1 size-3" />}
237 + label="Translate"
238 + tooltip="Can write English text from speech in another language. Turn on Translate to English to use it."
239 + />
240 + ) : null}
241 + </div>
242 + {model.description ? (
243 + <p className="mt-0.5 text-xs leading-normal text-muted-foreground">
244 + {model.description}
245 + </p>
246 ) : null}
205 - {model.supportsStreaming ? (
206 - <ModelBadge
207 - icon={<AudioLinesIcon className="mr-1 size-3" />}
208 - label="Live preview"
209 - tooltip="Shows words in your draft as you speak, before the recording finishes."
210 - />
247 + {model.linkedFrom ? (
248 + <p className="mt-0.5 break-all text-2xs text-muted-foreground">
249 + Using your file at {model.linkedFrom}
250 + </p>
251 ) : null}
212 - {model.supportsTranslation ? (
213 - <ModelBadge
214 - icon={<LanguagesIcon className="mr-1 size-3" />}
215 - label="Translate"
216 - tooltip="Can write English text from speech in another language. Turn on Translate to English to use it."
217 - />
252 + <ModelBenchmarks
253 + model={model}
254 + languages={props.comparisonLanguages}
255 + tier={props.benchmarkTier}
256 + facts={[
257 + ...(model.languages.length > 1
258 + ? [`${model.languages.length} languages`]
259 + : model.benchmark
260 + ? []
261 + : [languageLabel(model.languages[0] ?? "")]),
262 + formatSize(model.size),
263 + ]}
264 + />
265 + {downloading ? (
266 + <div className="mt-2">
267 + <div
268 + role="progressbar"
269 + aria-label={`${model.name} download progress`}
270 + aria-valuemin={0}
271 + aria-valuemax={100}
272 + aria-valuenow={Math.round(progress)}
273 + className="h-1 overflow-hidden rounded-full bg-muted"
274 + >
275 + <div
276 + className="h-full bg-primary transition-[width] duration-200"
277 + style={{ width: `${progress}%` }}
278 + />
279 + </div>
280 + <p className="mt-1 text-2xs text-muted-foreground">
281 + {model.state === "verifying"
282 + ? "Verifying download…"
283 + : `${Math.round(progress)}% downloaded`}
284 + </p>
285 + </div>
286 ) : null}
287 </div>
220 - <p className="mt-0.5 text-xs leading-relaxed text-muted-foreground">{model.description}</p>
221 - {model.linkedFrom ? (
222 - <p className="mt-0.5 break-all text-2xs text-muted-foreground">
223 - Using your file at {model.linkedFrom}
224 - </p>
225 - ) : null}
226 - <ModelBenchmarks
227 - model={model}
228 - languages={props.comparisonLanguages}
229 - tier={props.benchmarkTier}
230 - facts={[
231 - ...(model.languages.length > 1
232 - ? [`${model.languages.length} languages`]
233 - : model.benchmark
234 - ? []
235 - : [languageLabel(model.languages[0] ?? "")]),
236 - formatSize(model.size),
237 - ]}
238 - />
239 - {downloading ? (
240 - <div className="mt-2">
241 - <div
242 - role="progressbar"
243 - aria-label={`${model.name} download progress`}
244 - aria-valuemin={0}
245 - aria-valuemax={100}
246 - aria-valuenow={Math.round(progress)}
247 - className="h-1 overflow-hidden rounded-full bg-muted"
288 + <div className="flex max-w-full shrink-0 flex-wrap items-center gap-1.5">
289 + {props.testControls}
290 + {downloading ? (
291 + <Button
292 + size="icon-sm"
293 + variant="outline"
294 + aria-label={`Cancel ${model.name} download`}
295 + disabled={!props.canCancel}
296 + onClick={props.onCancel}
297 >
249 - <div
250 - className="h-full bg-primary transition-[width] duration-200"
251 - style={{ width: `${progress}%` }}
252 - />
253 - </div>
254 - <p className="mt-1 text-2xs text-muted-foreground">
255 - {model.state === "verifying"
256 - ? "Verifying download…"
257 - : `${Math.round(progress)}% downloaded`}
258 - </p>
259 - </div>
260 - ) : null}
261 - </div>
262 - <div className="flex max-w-full flex-wrap items-center justify-end gap-1">
263 - {props.testControls}
264 - {downloading ? (
265 - <Button
266 - size="icon-sm"
267 - variant="ghost"
268 - aria-label={`Cancel ${model.name} download`}
269 - disabled={!props.canCancel}
270 - onClick={props.onCancel}
271 - >
272 - <XIcon className="size-3.5" />
273 - </Button>
274 - ) : model.state === "downloadable" ? (
275 - <Button size="sm" variant="outline" disabled={props.busy} onClick={props.onDownload}>
276 - <DownloadIcon className="mr-1.5 size-3.5" />
277 - Download
278 - </Button>
279 - ) : model.active ? (
280 - <Button size="sm" variant="secondary" disabled>
281 - <CheckIcon className="mr-1.5 size-3.5" />
282 - Selected
283 - </Button>
284 - ) : (
285 - <Button size="sm" variant="outline" disabled={props.busy} onClick={props.onSelect}>
286 - Use model
287 - </Button>
288 - )}
289 - {model.state === "installed" ? (
290 - <Button
291 - size="icon-sm"
292 - variant="ghost"
293 - disabled={props.busy}
294 - aria-label={
295 - model.linkedFrom
296 - ? `Stop using your ${model.name} file; the file stays on disk`
297 - : `Delete ${model.name}`
298 - }
299 - onClick={props.onDelete}
300 - >
301 - {model.linkedFrom ? (
302 - <UnlinkIcon className="size-3.5" />
303 - ) : (
304 - <Trash2Icon className="size-3.5" />
305 - )}
306 - </Button>
307 - ) : null}
298 + <XIcon className="size-3.5" />
299 + </Button>
300 + ) : model.state === "downloadable" ? (
301 + <Button size="sm" variant="outline" disabled={props.busy} onClick={props.onDownload}>
302 + <DownloadIcon />
303 + Download
304 + </Button>
305 + ) : model.active ? (
306 + <Button size="sm" variant="secondary" disabled>
307 + <CheckIcon />
308 + Selected
309 + </Button>
310 + ) : (
311 + <Button size="sm" variant="outline" disabled={props.busy} onClick={props.onSelect}>
312 + Use model
313 + </Button>
314 + )}
315 + {model.state === "installed" ? (
316 + <Button
317 + size="icon-sm"
318 + variant="outline"
319 + disabled={props.busy}
320 + aria-label={
321 + model.linkedFrom
322 + ? `Stop using your ${model.name} file; the file stays on disk`
323 + : `Delete ${model.name}`
324 + }
325 + onClick={props.onDelete}
326 + >
327 + {model.linkedFrom ? <UnlinkIcon /> : <Trash2Icon />}
328 + </Button>
329 + ) : null}
330 + </div>
331 </div>
332 {props.children}
333 </div>
@@ -356,27 +379,34 @@ function DictionaryEditor(props: {
379 };
380 return (
381 <>
359 - <div className="flex items-start gap-1.5">
360 - <Textarea
361 - value={termDraft}
362 - rows={1}
363 - maxLength={5_100}
364 - disabled={disabled}
365 - placeholder="Add words or phrases, separated by commas or line breaks"
366 - aria-label={`Add a word or phrase to the ${label}`}
367 - onChange={(event) => setTermDraft(event.target.value)}
368 - onKeyDown={(event) => {
369 - if (event.key !== "Enter" || event.shiftKey) return;
370 - event.preventDefault();
371 - addTerm();
372 - }}
373 - />
374 - <Button type="button" disabled={!canAddTerm} onClick={addTerm}>
382 + <div className="flex items-start gap-2">
383 + {/* One line while empty, so a wrapped placeholder never makes the field taller than Add. */}
384 + <div className="min-w-0 flex-1 [&_textarea]:placeholder-shown:whitespace-nowrap">
385 + <Textarea
386 + value={termDraft}
387 + rows={1}
388 + maxLength={5_100}
389 + disabled={disabled}
390 + placeholder="Add words, separated by commas"
391 + aria-label={`Add a word or phrase to the ${label}`}
392 + onChange={(event) => setTermDraft(event.target.value)}
393 + onKeyDown={(event) => {
394 + if (event.key !== "Enter" || event.shiftKey) return;
395 + event.preventDefault();
396 + addTerm();
397 + }}
398 + />
399 + </div>
400 + <Button type="button" variant="outline" disabled={!canAddTerm} onClick={addTerm}>
401 + <PlusIcon />
402 Add
403 </Button>
404 </div>
405 {words.length > 0 ? (
379 - <ul aria-label={`${label} entries`} className="mt-3 flex flex-wrap gap-1.5">
406 + <ul
407 + aria-label={`${label} entries`}
408 + className="mt-2.5 flex flex-col gap-1.5 @min-[32rem]/settings-row:flex-row @min-[32rem]/settings-row:flex-wrap"
409 + >
410 {words.map(({ term, aliases }) => (
411 <li
412 key={term}
@@ -393,7 +423,7 @@ function DictionaryEditor(props: {
423 maxLength={50}
424 aria-label={`Alias for ${term}`}
425 placeholder="Misspelling"
396 - className="w-32"
426 + className="ms-auto w-32"
427 onChange={(event) => setAliasDraft({ term, value: event.target.value })}
428 onBlur={() => setAliasDraft(null)}
429 onKeyDown={(event) => {
@@ -408,6 +438,7 @@ function DictionaryEditor(props: {
438 type="button"
439 size="xs"
440 variant="ghost-muted"
441 + className="ms-auto"
442 disabled={disabled || aliases.length >= 8}
443 aria-label={`Add an alias for ${term}`}
444 onClick={() => setAliasDraft({ term, value: "" })}
@@ -448,16 +479,11 @@ function ProjectDictionaryRow() {
479 title={isProjectScope ? "Project dictionary" : "Shared dictionary"}
480 description={
481 isProjectScope
451 - ? `Saved with ${scope.scope.label} on its environment. Everyone who dictates in this project gets these spellings, along with their own dictionary. Project spellings win when the two overlap.`
452 - : "Saved on the environment and used in every project there, along with each person’s own dictionary. Pick a project above to limit these words to one project."
453 - }
454 - control={
455 - <span className="text-xs text-muted-foreground">
456 - {words.length} {words.length === 1 ? "entry" : "entries"}
457 - </span>
482 + ? `Used by everyone who dictates in ${scope.scope.label}.`
483 + : "Used by everyone who dictates on this environment."
484 }
485 >
460 - <div className="mt-2 pb-3.5">
486 + <div className="mt-3 pb-2">
487 <DictionaryEditor
488 label={isProjectScope ? "project dictionary" : "shared dictionary"}
489 words={words}
@@ -530,7 +556,7 @@ function CleanupModelRow() {
556 connectedEnvironments.length === 0 ? (
557 <span className="text-sm text-muted-foreground">Connect an environment first.</span>
558 ) : (
533 - <div className="flex min-w-0 flex-wrap items-center justify-end gap-2">
559 + <div className="flex min-w-0 flex-wrap items-center gap-2 @min-[32rem]/settings-row:justify-end">
560 {dedicated && (!hasProvider || !instanceEntry) ? (
561 <span className="text-sm text-muted-foreground">
562 No text generation providers available.
@@ -812,10 +838,14 @@ function SpeechModelBrowser(props: {
838 const accuracyLanguage = selectedLanguages[0] ?? "en";
839 const sortLabels: Record<SpeechModelSort, string> = {
840 recommended: "Recommended",
815 - size: "Smallest download",
816 - accuracy: `Best ${languageLabel(accuracyLanguage)} accuracy`,
817 - speed: "Fastest transcription",
841 + size: "Smallest",
842 + accuracy: "Most accurate",
843 + speed: "Fastest",
844 };
845 + const languagesLabel =
846 + selectedLanguages.length === 0
847 + ? "All languages"
848 + : selectedLanguages.map(languageLabel).join(", ");
849 const sortedModels = filterSpeechModels(models, {
850 search,
851 languages: selectedLanguages,
@@ -830,212 +860,204 @@ function SpeechModelBrowser(props: {
860 ? [activeModel, ...sortedModels]
861 : sortedModels;
862
863 + if (!status) return null;
864 + if (!status.supported)
865 + return (
866 + <SettingsRow
867 + {...searchableSetting("speech-models")}
868 + title="Models"
869 + description={`This environment can’t transcribe audio: ${status.reason}.`}
870 + />
871 + );
872 return (
834 - <SettingsSearchTarget
873 + <SettingsRow
874 {...searchableSetting("speech-models")}
875 + title="Models"
876 + description={`Grades are error rates and speed is seconds per 30 s of audio, from Handy’s benchmark ${status.benchmarkTier === "cpu" ? "CPU" : "GPU"}.`}
877 className="@container/voice-models"
878 >
838 - {status && !status.supported ? (
839 - <p className="px-4 py-2 text-sm text-muted-foreground">
840 - This environment can’t transcribe audio: {status.reason}.
841 - </p>
842 - ) : null}
843 - {status?.supported ? (
844 - <>
845 - <p className="border-b border-border/60 px-3 py-2 text-xs text-muted-foreground sm:px-4">
846 - Pick languages to see models that support all of them. Accuracy is word or character
847 - error rate: lower is better. Speed is the time to transcribe 30 seconds of audio on
848 - Handy’s benchmark {status.benchmarkTier === "cpu" ? "CPU" : "GPU"}.
849 - </p>
850 - <div className="@min-[32rem]/voice-models:grid @min-[32rem]/voice-models:grid-cols-[11rem_minmax(0,1fr)]">
851 - <div className="flex min-h-0 flex-col border-b border-border/60 bg-muted/10 @min-[32rem]/voice-models:border-r @min-[32rem]/voice-models:border-b-0">
852 - <div className="space-y-2 p-3">
853 - <p className="text-sm font-medium">Languages</p>
854 - <Input
855 - size="compact"
856 - aria-label="Search languages"
857 - placeholder="Search languages"
858 - value={languageSearch}
859 - onChange={(e) => setLanguageSearch(e.target.value)}
860 - />
861 - </div>
862 - <div
863 - role="group"
864 - aria-label="Browse models by language"
865 - className="flex gap-1 overflow-x-auto px-2 pb-2 @min-[32rem]/voice-models:h-72 @min-[32rem]/voice-models:flex-col @min-[32rem]/voice-models:overflow-y-auto"
866 - >
867 - <LanguageRow
868 - label="All languages"
869 - selected={selectedLanguages.length === 0}
870 - onClick={() => setSelection([])}
871 - />
872 - {visibleLanguages.map((code) => (
873 - <LanguageRow
874 - key={code}
875 - label={languageLabel(code)}
876 - selected={selectedLanguages.includes(code)}
877 - onClick={() =>
878 - setSelection(
879 - selectedLanguages.includes(code)
880 - ? selectedLanguages.filter((language) => language !== code)
881 - : [...selectedLanguages, code],
882 - )
883 - }
884 - />
885 - ))}
886 - {visibleLanguages.length === 0 && (
887 - <p className="px-2 py-1 text-xs text-muted-foreground">No languages match.</p>
888 - )}
889 - </div>
890 - </div>
891 - <div className="flex min-w-0 flex-col">
892 - <div className="space-y-2 p-3">
893 - <p className="text-sm font-medium">
894 - {selectedLanguages.length
895 - ? "Models for " + selectedLanguages.map(languageLabel).join(", ")
896 - : "All models"}
897 - </p>
898 - <InputGroup>
899 - <InputGroupInput
900 - size="compact"
901 - aria-label="Search transcription models"
902 - placeholder="Search models"
903 - value={search}
904 - onChange={(e) => setSearch(e.target.value)}
905 - />
906 - <InputGroupAddon align="inline-end">
907 - <Menu>
908 - <MenuTrigger
909 - render={<Button size="compact" variant="ghost-muted" />}
910 - aria-label={`Sort and filter models: ${sortLabels[sort]}${downloadedOnly ? ", downloaded only" : ""}`}
911 - >
912 - <SlidersHorizontalIcon />
913 - <span className="hidden @min-[40rem]/voice-models:inline">
914 - {sortLabels[sort]}
915 - </span>
916 - {downloadedOnly && <span>1</span>}
917 - </MenuTrigger>
918 - <MenuPopup align="end">
919 - <MenuGroup>
920 - <MenuGroupLabel>Sort by</MenuGroupLabel>
921 - <MenuRadioGroup
922 - value={sort}
923 - onValueChange={(value) => setSort(value as SpeechModelSort)}
924 - >
925 - {(Object.keys(sortLabels) as SpeechModelSort[]).map((value) => (
926 - <MenuRadioItem key={value} value={value} closeOnClick>
927 - <span className="flex items-center justify-between gap-3">
928 - {sortLabels[value]}
929 - <MenuRadioItemIndicator />
930 - </span>
931 - </MenuRadioItem>
932 - ))}
933 - </MenuRadioGroup>
934 - </MenuGroup>
935 - <MenuSeparator />
936 - <MenuCheckboxItem
937 - checked={downloadedOnly}
938 - onCheckedChange={setDownloadedOnly}
939 - >
940 - Downloaded only
941 - </MenuCheckboxItem>
942 - </MenuPopup>
943 - </Menu>
944 - </InputGroupAddon>
945 - </InputGroup>
946 - </div>
947 - <div
948 - className="h-72 overflow-y-auto divide-y divide-border/60"
949 - role="region"
950 - aria-label="Transcription models"
951 - tabIndex={0}
952 - >
953 - {sortedModels.length === 0 && (
954 - <p className="px-4 py-2 text-sm text-muted-foreground">No models match.</p>
955 - )}
956 - {renderedModels.map((model) => {
957 - const card: ModelCardProps = {
958 - model:
959 - model.id === requestedDownload && model.state === "downloadable"
960 - ? { ...model, state: "downloading" }
961 - : model,
962 - busy: controlsDisabled,
963 - canCancel,
964 - comparisonLanguages: selectedLanguages,
965 - benchmarkTier: status.benchmarkTier,
966 - onDownload: () =>
967 - void run(
968 - "Could not download the model",
969 - () =>
970 - runEnvironmentSpeech(downloadEnvironmentSpeechModel(prepared, model.id)),
971 - model.id,
972 - ),
973 - onSelect: () =>
974 - void run("Could not select the model", () =>
975 - runEnvironmentSpeech(selectEnvironmentSpeechModel(prepared, model.id)),
976 - ),
977 - onCancel: () =>
978 - void run("Could not cancel the download", () =>
979 - runEnvironmentSpeech(
980 - cancelEnvironmentSpeechModelDownload(prepared, model.id),
981 - ),
982 - ),
983 - onDelete: () =>
984 - void run("Could not delete the model", () =>
985 - runEnvironmentSpeech(removeEnvironmentSpeechModel(prepared, model.id)),
986 - ),
987 - };
988 - return model.active ? (
989 - <TranscriptionTest
990 - key={model.id}
991 - card={card}
992 - hidden={!sortedModels.includes(model)}
993 - environmentId={props.testEnvironmentId}
994 - projectId={props.projectId}
995 - projectName={props.projectName}
996 - />
997 - ) : (
998 - <ModelCard key={model.id} {...card} />
999 - );
1000 - })}
1001 - </div>
1002 - {status.supportsModelLinking && canReadFiles ? (
1003 - <form
1004 - className="flex items-center gap-2 border-t border-border/60 p-3"
1005 - onSubmit={(event) => {
1006 - event.preventDefault();
1007 - void run("Could not use a model from that path", async () => {
1008 - await runEnvironmentSpeech(
1009 - linkEnvironmentSpeechModels(prepared, existingPath),
1010 - );
1011 - setExistingPath("");
1012 - });
1013 - }}
879 + <div className="mt-3 flex min-w-0 gap-2">
880 + <InputGroup className="min-w-0 flex-1 **:[input]:h-9 sm:**:[input]:h-8">
881 + <InputGroupAddon>
882 + <SearchIcon aria-hidden />
883 + </InputGroupAddon>
884 + <InputGroupInput
885 + type="search"
886 + aria-label="Search transcription models"
887 + placeholder="Search models"
888 + value={search}
889 + onChange={(e) => setSearch(e.target.value)}
890 + />
891 + </InputGroup>
892 + <Menu>
893 + <MenuTrigger
894 + render={<Button variant="outline" className="max-w-56" />}
895 + aria-label={`Filter by language: ${languagesLabel}`}
896 + >
897 + <LanguagesIcon />
898 + <span className="hidden truncate @min-[34rem]/voice-models:inline">
899 + {languagesLabel}
900 + </span>
901 + </MenuTrigger>
902 + <MenuPopup align="end">
903 + <MenuSearchInput
904 + autoFocus
905 + aria-label="Search languages"
906 + placeholder="Search languages"
907 + value={languageSearch}
908 + onChange={(e) => setLanguageSearch(e.target.value)}
909 + />
910 + <div
911 + role="group"
912 + aria-label="Browse models by language"
913 + className="max-h-72 overflow-y-auto"
914 + >
915 + {needle ? null : (
916 + <MenuCheckboxItem
917 + checked={selectedLanguages.length === 0}
918 + onCheckedChange={() => setSelection([])}
919 >
1015 - <div className="min-w-0 flex-1">
1016 - <Input
1017 - size="compact"
1018 - aria-label="Path to a model file or folder already on this environment"
1019 - placeholder="Already downloaded elsewhere? Paste the model file or folder path"
1020 - value={existingPath}
1021 - onChange={(e) => setExistingPath(e.target.value)}
1022 - />
1023 - </div>
1024 - <Button
1025 - type="submit"
1026 - size="sm"
1027 - variant="outline"
1028 - disabled={controlsDisabled || !existingPath.trim()}
1029 - >
1030 - Use existing
1031 - </Button>
1032 - </form>
1033 - ) : null}
920 + All languages
921 + </MenuCheckboxItem>
922 + )}
923 + {visibleLanguages.map((code) => (
924 + <MenuCheckboxItem
925 + key={code}
926 + checked={selectedLanguages.includes(code)}
927 + onCheckedChange={() =>
928 + setSelection(
929 + selectedLanguages.includes(code)
930 + ? selectedLanguages.filter((language) => language !== code)
931 + : [...selectedLanguages, code],
932 + )
933 + }
934 + >
935 + {languageLabel(code)}
936 + </MenuCheckboxItem>
937 + ))}
938 + {visibleLanguages.length === 0 && (
939 + <p className="px-2 py-1.5 text-sm text-muted-foreground">No languages match.</p>
940 + )}
941 </div>
942 + </MenuPopup>
943 + </Menu>
944 + <Menu>
945 + <MenuTrigger
946 + render={<Button variant="outline" />}
947 + aria-label={`Sort and filter models: ${sortLabels[sort]}${downloadedOnly ? ", downloaded only" : ""}`}
948 + >
949 + <ArrowDownUpIcon />
950 + <span className="hidden @min-[34rem]/voice-models:inline">{sortLabels[sort]}</span>
951 + </MenuTrigger>
952 + <MenuPopup align="end">
953 + <MenuGroup>
954 + <MenuGroupLabel>Sort by</MenuGroupLabel>
955 + <MenuRadioGroup
956 + value={sort}
957 + onValueChange={(value) => setSort(value as SpeechModelSort)}
958 + >
959 + {(Object.keys(sortLabels) as SpeechModelSort[]).map((value) => (
960 + <MenuRadioItem key={value} value={value} closeOnClick>
961 + <span className="flex items-center justify-between gap-3">
962 + {value === "accuracy"
963 + ? `Most accurate in ${languageLabel(accuracyLanguage)}`
964 + : sortLabels[value]}
965 + <MenuRadioItemIndicator />
966 + </span>
967 + </MenuRadioItem>
968 + ))}
969 + </MenuRadioGroup>
970 + </MenuGroup>
971 + <MenuSeparator />
972 + <MenuCheckboxItem checked={downloadedOnly} onCheckedChange={setDownloadedOnly}>
973 + Downloaded only
974 + </MenuCheckboxItem>
975 + </MenuPopup>
976 + </Menu>
977 + </div>
978 + <div
979 + className="mt-2 mb-2 max-h-96 divide-y divide-border/60 overflow-y-auto rounded-lg border border-border/60"
980 + role="region"
981 + aria-label="Transcription models"
982 + tabIndex={0}
983 + >
984 + {sortedModels.length === 0 && (
985 + <p className="px-3 py-3 text-sm text-muted-foreground">No models match.</p>
986 + )}
987 + {renderedModels.map((model) => {
988 + const card: ModelCardProps = {
989 + model:
990 + model.id === requestedDownload && model.state === "downloadable"
991 + ? { ...model, state: "downloading" }
992 + : model,
993 + busy: controlsDisabled,
994 + canCancel,
995 + comparisonLanguages: selectedLanguages,
996 + benchmarkTier: status.benchmarkTier,
997 + onDownload: () =>
998 + void run(
999 + "Could not download the model",
1000 + () => runEnvironmentSpeech(downloadEnvironmentSpeechModel(prepared, model.id)),
1001 + model.id,
1002 + ),
1003 + onSelect: () =>
1004 + void run("Could not select the model", () =>
1005 + runEnvironmentSpeech(selectEnvironmentSpeechModel(prepared, model.id)),
1006 + ),
1007 + onCancel: () =>
1008 + void run("Could not cancel the download", () =>
1009 + runEnvironmentSpeech(cancelEnvironmentSpeechModelDownload(prepared, model.id)),
1010 + ),
1011 + onDelete: () =>
1012 + void run("Could not delete the model", () =>
1013 + runEnvironmentSpeech(removeEnvironmentSpeechModel(prepared, model.id)),
1014 + ),
1015 + };
1016 + return model.active ? (
1017 + <TranscriptionTest
1018 + key={model.id}
1019 + card={card}
1020 + hidden={!sortedModels.includes(model)}
1021 + environmentId={props.testEnvironmentId}
1022 + projectId={props.projectId}
1023 + projectName={props.projectName}
1024 + />
1025 + ) : (
1026 + <ModelCard key={model.id} {...card} />
1027 + );
1028 + })}
1029 + </div>
1030 + {status.supportsModelLinking && canReadFiles ? (
1031 + <form
1032 + className="mb-2 flex flex-col items-start gap-2 @min-[28rem]/voice-models:flex-row @min-[28rem]/voice-models:items-center"
1033 + onSubmit={(event) => {
1034 + event.preventDefault();
1035 + void run("Could not use a model from that path", async () => {
1036 + await runEnvironmentSpeech(linkEnvironmentSpeechModels(prepared, existingPath));
1037 + setExistingPath("");
1038 + });
1039 + }}
1040 + >
1041 + <div className="w-full min-w-0 flex-1">
1042 + <Input
1043 + size="compact"
1044 + aria-label="Path to a model file or folder already on this environment"
1045 + placeholder="Already downloaded elsewhere? Paste the model file or folder path"
1046 + value={existingPath}
1047 + onChange={(e) => setExistingPath(e.target.value)}
1048 + />
1049 </div>
1036 - </>
1050 + <Button
1051 + type="submit"
1052 + size="sm"
1053 + variant="outline"
1054 + disabled={controlsDisabled || !existingPath.trim()}
1055 + >
1056 + Use existing
1057 + </Button>
1058 + </form>
1059 ) : null}
1038 - </SettingsSearchTarget>
1060 + </SettingsRow>
1061 );
1062 }
1063
@@ -1071,8 +1093,9 @@ function TranscriptionTest(props: {
1093 <SettingsSearchTarget {...searchableSetting("transcription-test")} hidden={props.hidden}>
1094 <ModelCard
1095 {...props.card}
1096 + stacked={showTranscript}
1097 testControls={
1075 - <div className="flex min-w-0 max-w-full items-center gap-1">
1098 + <div className="order-last flex min-w-0 max-w-full basis-full items-center gap-1 @min-[28rem]/model:order-none @min-[28rem]/model:basis-auto">
1099 <ComposerSpeechRecordingPill
1100 state={speech.state}
1101 progress={null}
@@ -1107,40 +1130,41 @@ function TranscriptionTest(props: {
1130 }
1131 >
1132 {showTranscript ? (
1110 - <div className="w-full">
1111 - <div className="flex items-start gap-2">
1112 - <p className="min-w-0 flex-1 text-xs leading-relaxed text-muted-foreground">
1113 - Say a sentence in the language above. Cleanup is skipped, so you see exactly what
1114 - the model heard. The transcript stays on this page.
1115 - </p>
1116 - <Button
1117 - size="icon-sm"
1118 - variant="ghost"
1119 - aria-label="Close model test"
1120 - onClick={() => {
1121 - speech.cancel();
1122 - setShowTranscript(false);
1123 - }}
1124 - >
1125 - <XIcon />
1126 - </Button>
1133 + <div className="mt-3 flex items-start gap-3 rounded-lg border border-border/60 bg-muted/30 py-1 ps-3 pe-1">
1134 + <div className="min-w-0 flex-1 py-1.5">
1135 + {speech.preview || transcript !== null ? null : (
1136 + <p className="text-xs text-muted-foreground">
1137 + Say a sentence. Cleanup is skipped, so you see exactly what the model heard.
1138 + </p>
1139 + )}
1140 + {speech.preview || transcript !== null ? (
1141 + <p
1142 + role="status"
1143 + aria-label="Test transcript"
1144 + className="select-text whitespace-pre-wrap break-words text-sm leading-relaxed"
1145 + >
1146 + {speech.preview ? (
1147 + <>
1148 + {speech.preview.committed}
1149 + <span className="text-muted-foreground">{speech.preview.tentative}</span>
1150 + </>
1151 + ) : (
1152 + transcript || "No speech recognized. Try again."
1153 + )}
1154 + </p>
1155 + ) : null}
1156 </div>
1128 - {speech.preview || transcript !== null ? (
1129 - <p
1130 - role="status"
1131 - aria-label="Test transcript"
1132 - className="mt-2 select-text whitespace-pre-wrap break-words text-sm leading-relaxed"
1133 - >
1134 - {speech.preview ? (
1135 - <>
1136 - {speech.preview.committed}
1137 - <span className="text-muted-foreground">{speech.preview.tentative}</span>
1138 - </>
1139 - ) : (
1140 - transcript || "No speech recognized. Try again."
1141 - )}
1142 - </p>
1143 - ) : null}
1157 + <Button
1158 + size="icon-sm"
1159 + variant="ghost-muted"
1160 + aria-label="Close model test"
1161 + onClick={() => {
1162 + speech.cancel();
1163 + setShowTranscript(false);
1164 + }}
1165 + >
1166 + <XIcon />
1167 + </Button>
1168 </div>
1169 ) : null}
1170 </ModelCard>
@@ -1181,7 +1205,9 @@ export function VoiceSettingsPanel() {
1205 const refreshMicrophones = useStableCallback(async () => {
1206 if (!navigator.mediaDevices?.enumerateDevices) return;
1207 try {
1184 - setMicrophones(await listBrowserMicrophones());
1208 + const devices = await listAudioInputs();
1209 + setMicrophones(devices);
1210 + updateMicrophoneOrder((order) => rememberMicrophones(order, devices));
1211 } catch {
1212 setMicrophones([]);
1213 }
@@ -1203,11 +1229,8 @@ export function VoiceSettingsPanel() {
1229 (member) => member.environmentId === target?.environmentId && member.id === target.projectId,
1230 ) ?? members?.[0];
1231
1206 - const selectedMicrophone = settings.voiceMicrophone;
1207 - const selectedMicrophoneLabel = selectedMicrophone
1208 - ? (microphones.find((device) => device.deviceId === selectedMicrophone)?.label ??
1209 - "Selected microphone (unavailable)")
1210 - : "System default";
1232 + const order = microphoneOrder(settings);
1233 + const selectedMicrophone = preferredMicrophoneId(order, microphones);
1234 const environmentChoices = environments.map((environment) => ({
1235 environment,
1236 value: JSON.stringify(["environment", environment.environmentId]),
@@ -1266,52 +1289,39 @@ export function VoiceSettingsPanel() {
1289 <SettingsRow
1290 {...searchableSetting("voice-microphone")}
1291 title="Microphone"
1269 - description="System default follows your operating system. Record a ten-second test and play it back here. Test audio stays on this device and is never transcribed."
1292 + description="Voice input uses the first connected microphone in this list."
1293 control={
1271 - <div className="flex min-w-0 flex-wrap items-center justify-end gap-1.5">
1272 - <SettingsSearchTarget {...searchableSetting("microphone-test")}>
1273 - <MicrophoneTest key={selectedMicrophone} microphoneId={selectedMicrophone} />
1274 - </SettingsSearchTarget>
1275 - <div className="flex min-w-0 items-center gap-1.5">
1276 - <Select
1277 - value={selectedMicrophone ? `device:${selectedMicrophone}` : SYSTEM_DEFAULT}
1278 - onValueChange={(value) => {
1279 - if (value)
1280 - void updateClientSettings({
1281 - voiceMicrophone:
1282 - value === SYSTEM_DEFAULT ? "" : value.slice("device:".length),
1283 - });
1284 - }}
1285 - >
1286 - <SelectTrigger size="sm" aria-label="Microphone" className="max-w-64">
1287 - <SelectValue>{selectedMicrophoneLabel}</SelectValue>
1288 - </SelectTrigger>
1289 - <SelectPopup align="end" alignItemWithTrigger={false}>
1290 - <SelectItem value={SYSTEM_DEFAULT}>System default</SelectItem>
1291 - {microphones.map((device) => (
1292 - <SelectItem key={device.deviceId} value={`device:${device.deviceId}`}>
1293 - {device.label}
1294 - </SelectItem>
1295 - ))}
1296 - </SelectPopup>
1297 - </Select>
1298 - <Button
1299 - type="button"
1300 - variant="ghost"
1301 - size="icon-sm"
1302 - aria-label="Refresh microphones"
1303 - onClick={() => void refreshMicrophones()}
1304 - >
1305 - <RefreshCwIcon className="size-3.5" />
1306 - </Button>
1307 - </div>
1308 - </div>
1294 + <Tooltip>
1295 + <TooltipTrigger
1296 + render={
1297 + <Button
1298 + type="button"
1299 + variant="outline"
1300 + size="icon-sm"
1301 + aria-label="Refresh microphones"
1302 + onClick={() => void refreshMicrophones()}
1303 + >
1304 + <RefreshCwIcon />
1305 + </Button>
1306 + }
1307 + />
1308 + <TooltipPopup>Refresh microphones</TooltipPopup>
1309 + </Tooltip>
1310 }
1311 + >
1312 + <div className="mt-3 pb-2">
1313 + <MicrophoneOrderList order={order} devices={microphones} />
1314 + </div>
1315 + </SettingsRow>
1316 + <MicrophoneTest
1317 + key={selectedMicrophone}
1318 + microphoneId={selectedMicrophone}
1319 + onMicrophoneAccess={() => void refreshMicrophones()}
1320 />
1321 <SettingsRow
1322 {...searchableSetting("dictation-shortcut-mode")}
1323 title="Dictation shortcut"
1314 - description="How the dictation shortcut starts and stops recording."
1324 + description={`${SHORTCUT_MODES[settings.voiceShortcutMode].hint}.`}
1325 control={
1326 <Select
1327 value={settings.voiceShortcutMode}
@@ -1320,13 +1330,18 @@ export function VoiceSettingsPanel() {
1330 void updateClientSettings({ voiceShortcutMode: value as VoiceShortcutMode });
1331 }}
1332 >
1323 - <SelectTrigger size="sm" aria-label="Dictation shortcut mode" className="max-w-72">
1324 - <SelectValue>{SHORTCUT_MODE_LABELS[settings.voiceShortcutMode]}</SelectValue>
1333 + <SelectTrigger size="sm" aria-label="Dictation shortcut mode">
1334 + <SelectValue>{SHORTCUT_MODES[settings.voiceShortcutMode].label}</SelectValue>
1335 </SelectTrigger>
1336 <SelectPopup align="end" alignItemWithTrigger={false}>
1327 - {(Object.keys(SHORTCUT_MODE_LABELS) as VoiceShortcutMode[]).map((mode) => (
1337 + {(Object.keys(SHORTCUT_MODES) as VoiceShortcutMode[]).map((mode) => (
1338 <SelectItem key={mode} value={mode}>
1329 - {SHORTCUT_MODE_LABELS[mode]}
1339 + <span className="flex flex-col">
1340 + {SHORTCUT_MODES[mode].label}
1341 + <span className="text-xs text-muted-foreground">
1342 + {SHORTCUT_MODES[mode].hint}
1343 + </span>
1344 + </span>
1345 </SelectItem>
1346 ))}
1347 </SelectPopup>
@@ -1336,16 +1351,16 @@ export function VoiceSettingsPanel() {
1351 </SettingsSection>
1352
1353 <SettingsSection id="voice-transcription" title="Transcription">
1339 - <SettingsSearchTarget
1354 + <SettingsRow
1355 {...searchableSetting("voice-transcription-destination")}
1341 - className="@container/destination px-3 py-3 sm:px-4"
1342 - >
1343 - <div className="flex items-center justify-between gap-2">
1344 - <h3 className="shrink-0 text-sm font-medium">
1345 - <span className="@max-[18rem]/destination:hidden">Transcribe with</span>
1346 - <span className="@min-[18rem]/destination:hidden">Use</span>
1347 - </h3>
1348 - <div className="flex min-w-0 items-center justify-end gap-1">
1356 + title="Transcribe with"
1357 + description={
1358 + destination.environmentId && chosenUnsupported
1359 + ? `${chosenEnvironment.label} can’t transcribe audio locally. Choose another destination.`
1360 + : "Where your speech is turned into text."
1361 + }
1362 + control={
1363 + <div className="flex w-full min-w-0 items-center gap-1.5 @min-[32rem]/settings-row:w-auto">
1364 <Select
1365 value={destinationValue}
1366 onValueChange={(value) => {
@@ -1371,7 +1386,7 @@ export function VoiceSettingsPanel() {
1386 <SelectTrigger
1387 size="sm"
1388 aria-label="Transcribe with"
1374 - className="w-auto min-w-0 max-w-72"
1389 + className="min-w-0 @min-[32rem]/settings-row:w-auto @min-[32rem]/settings-row:max-w-72"
1390 >
1391 <SelectValue>
1392 <span className="flex min-w-0 items-center gap-2">
@@ -1472,15 +1487,8 @@ export function VoiceSettingsPanel() {
1487 </Tooltip>
1488 )}
1489 </div>
1475 - </div>
1476 - {destination.environmentId && chosenUnsupported ? (
1477 - <p className="mt-2 text-xs text-destructive">
1478 - {chosenEnvironment.label} can’t transcribe audio locally. Choose a different
1479 - destination.
1480 - </p>
1481 - ) : null}
1482 - </SettingsSearchTarget>
1483 -
1490 + }
1491 + />
1492 {pendingCloud && cloudEnvironment && (
1493 <CloudVoiceConfiguration
1494 key={JSON.stringify(pendingCloud)}
@@ -1499,10 +1507,10 @@ export function VoiceSettingsPanel() {
1507 title="Language"
1508 description={
1509 destination.cloudProvider
1502 - ? "The cloud provider can detect your language automatically."
1503 - : model
1504 - ? `${model.model} ${model.supportsLanguageDetection ? "detects the language automatically" : "uses one fixed language"}.`
1505 - : "Choose an environment above to see the languages its transcription model supports."
1510 + ? "The provider detects your language automatically."
1511 + : model?.supportsLanguageDetection
1512 + ? `${model.model} detects the language automatically.`
1513 + : `${model?.model ?? "This model"} uses one fixed language.`
1514 }
1515 control={
1516 <SpeechLanguageSelect
@@ -1526,28 +1534,13 @@ export function VoiceSettingsPanel() {
1534 }
1535 />
1536 ) : null}
1529 - {prepared ? (
1530 - <SpeechModelBrowser
1531 - key={prepared.environmentId}
1532 - prepared={prepared}
1533 - catalog={catalog}
1534 - testEnvironmentId={
1535 - project?.environmentId ?? target?.environmentId ?? prepared.environmentId
1536 - }
1537 - projectId={project?.id ?? null}
1538 - projectName={project?.title ?? null}
1539 - />
1540 - ) : null}
1541 - {prepared && <AdvancedSpeechSettings prepared={prepared} catalog={catalog} />}
1542 - </SettingsSection>
1543 - <SettingsSection id="voice-recognition" title="Transcription options">
1537 <SettingsRow
1538 {...searchableSetting("speech-silence-filtering")}
1539 title="Silence filtering"
1540 description={
1541 destination.cloudProvider === "openai"
1549 - ? "OpenAI handles silence automatically."
1550 - : "Skip silence and other non-speech audio before transcription, keeping a short buffer so words aren’t cut off."
1542 + ? "OpenAI handles silence itself."
1543 + : "Skip silence and background noise before transcribing."
1544 }
1545 control={
1546 <Switch
@@ -1563,7 +1556,7 @@ export function VoiceSettingsPanel() {
1556 <SettingsRow
1557 {...searchableSetting("remove-filler-words")}
1558 title="Remove filler words"
1566 - description="Remove filler words like “um” and “uh” from transcripts."
1559 + description="Drop “um” and “uh” from transcripts."
1560 control={
1561 <Switch
1562 aria-label="Remove filler words"
@@ -1574,18 +1567,28 @@ export function VoiceSettingsPanel() {
1567 />
1568 }
1569 />
1570 + {prepared ? (
1571 + <SpeechModelBrowser
1572 + key={prepared.environmentId}
1573 + prepared={prepared}
1574 + catalog={catalog}
1575 + testEnvironmentId={
1576 + project?.environmentId ?? target?.environmentId ?? prepared.environmentId
1577 + }
1578 + projectId={project?.id ?? null}
1579 + projectName={project?.title ?? null}
1580 + />
1581 + ) : null}
1582 + {prepared && <AdvancedSpeechSettings prepared={prepared} catalog={catalog} />}
1583 + </SettingsSection>
1584 +
1585 + <SettingsSection id="voice-dictionary" title="Dictionary">
1586 <SettingsRow
1587 {...searchableSetting("speech-dictionary")}
1579 - title="Dictionary"
1580 - description="Names and terms transcription should spell your way, saved on this device. Add an alias when it keeps hearing a term wrong. The name of the project you dictate in is also sent as a hint."
1581 - control={
1582 - <span className="text-xs text-muted-foreground">
1583 - {settings.speechCustomWords.length}{" "}
1584 - {settings.speechCustomWords.length === 1 ? "entry" : "entries"}
1585 - </span>
1586 - }
1588 + title="Your dictionary"
1589 + description="Names and terms to spell your way, saved on this device."
1590 >
1588 - <div className="mt-2 pb-3.5">
1591 + <div className="mt-3 pb-2">
1592 <DictionaryEditor
1593 label="dictionary"
1594 words={settings.speechCustomWords}
@@ -1599,7 +1602,7 @@ export function VoiceSettingsPanel() {
1602 <SettingsSection id={searchableSetting("speech-post-processing").id} title="Cleanup">
1603 <SettingsRow
1604 title="Clean up transcripts"
1602 - description="Fixes punctuation, numbers, and spoken corrections with the cleanup model on the draft’s environment. Sends the transcript and nearby draft text, never audio."
1605 + description="Off by default. Sends the transcript and nearby draft text, never audio, to the cleanup model."
1606 control={
1607 <Switch
1608 aria-label="Clean up transcripts"
@@ -1610,8 +1613,12 @@ export function VoiceSettingsPanel() {
1613 />
1614 {cleanupEnabled ? (
1615 <SettingsRow
1613 - title="Cleanup instructions"
1614 - description="Use the built-in instructions or write your own."
1616 + title="Instructions"
1617 + description={
1618 + settings.speechPostProcessingPrompt.mode === "custom"
1619 + ? "Your own instructions for cleanup."
1620 + : "Fix punctuation, numbers, lists, and spoken corrections."
1621 + }
1622 control={
1623 <Select
1624 value={settings.speechPostProcessingPrompt.mode}
@@ -1619,13 +1626,17 @@ export function VoiceSettingsPanel() {
1626 if (value === "default" || value === "custom")
1627 void updateClientSettings({
1628 speechPostProcessingPrompt: {
1622 - ...settings.speechPostProcessingPrompt,
1629 mode: value,
1630 + customInstructions:
1631 + value === "custom" &&
1632 + !settings.speechPostProcessingPrompt.customInstructions.trim()
1633 + ? DEFAULT_SPEECH_POST_PROCESSING_PROMPT
1634 + : settings.speechPostProcessingPrompt.customInstructions,
1635 },
1636 });
1637 }}
1638 >
1628 - <SelectTrigger size="sm" aria-label="Cleanup instructions" className="max-w-64">
1639 + <SelectTrigger size="sm" aria-label="Cleanup instructions">
1640 <SelectValue>
1641 {settings.speechPostProcessingPrompt.mode === "custom" ? "Custom" : "Built-in"}
1642 </SelectValue>
@@ -1637,25 +1648,23 @@ export function VoiceSettingsPanel() {
1648 </Select>
1649 }
1650 >
1640 - <Textarea
1641 - aria-label="Custom cleanup instructions"
1642 - className="mt-2"
1643 - maxLength={10_000}
1644 - disabled={settings.speechPostProcessingPrompt.mode !== "custom"}
1645 - value={
1646 - settings.speechPostProcessingPrompt.mode === "custom"
1647 - ? settings.speechPostProcessingPrompt.customInstructions
1648 - : DEFAULT_SPEECH_POST_PROCESSING_PROMPT
1649 - }
1650 - onChange={(event) =>
1651 - void updateClientSettings({
1652 - speechPostProcessingPrompt: {
1653 - mode: "custom",
1654 - customInstructions: event.target.value,
1655 - },
1656 - })
1657 - }
1658 - />
1651 + {settings.speechPostProcessingPrompt.mode === "custom" ? (
1652 + <div className="mt-3 pb-2">
1653 + <Textarea
1654 + aria-label="Custom cleanup instructions"
1655 + maxLength={10_000}
1656 + value={settings.speechPostProcessingPrompt.customInstructions}
1657 + onChange={(event) =>
1658 + void updateClientSettings({
1659 + speechPostProcessingPrompt: {
1660 + mode: "custom",
1661 + customInstructions: event.target.value,
1662 + },
1663 + })
1664 + }
1665 + />
1666 + </div>
1667 + ) : null}
1668 </SettingsRow>
1669 ) : null}
1670 {cleanupEnabled && settingsScope !== null ? <CleanupModelRow /> : null}
@@ -1663,7 +1672,7 @@ export function VoiceSettingsPanel() {
1672 <SettingsRow
1673 {...searchableSetting("speech-correction-word")}
1674 title="Correction cue"
1666 - description="The word or phrase you say to flag a correction, like “sorry” or “scratch that”. Cleanup applies the correction only where one is meant, and hears the cue as a hint."
1675 + description="The phrase you say to fix a mistake, like “scratch that”."
1676 control={
1677 <Input
1678 key={settings.speechCorrectionWord}
@@ -1671,7 +1680,7 @@ export function VoiceSettingsPanel() {
1680 maxLength={50}
1681 placeholder="sorry"
1682 aria-label="Correction cue"
1674 - className="w-40"
1683 + className="@min-[32rem]/settings-row:w-40"
1684 onBlur={(event) => {
1685 const next = event.target.value.trim();
1686 if (next !== settings.speechCorrectionWord)
apps/web/src/components/settings/settingsSearch.ts
+1 -1
@@ -540,7 +540,7 @@ export const SETTINGS_SEARCH_ITEMS = [
540 id: "voice-microphone",
541 title: "Microphone",
542 to: "/settings/voice",
543 - searchTerms: ["mic input audio device"],
543 + searchTerms: ["mic input audio device order priority preferred"],
544 },
545 {
546 id: "dictation-shortcut-mode",
apps/web/src/speech/browserMicrophones.ts
+43 -5
@@ -1,11 +1,49 @@
1 +import {
2 + getClientSettings,
3 + isClientSettingsHydrated,
4 + updateClientSettings,
5 +} from "../hooks/useSettings";
6 +import {
7 + type MicrophoneEntry,
8 + microphoneOrder,
9 + preferredMicrophoneId,
10 + rememberMicrophones,
11 +} from "./microphoneOrder";
12 +
13 export type BrowserMicrophone = Pick<MediaDeviceInfo, "deviceId" | "label">;
14
3 -export async function listBrowserMicrophones(): Promise<BrowserMicrophone[]> {
15 +/** Audio inputs as the browser reports them. Labels stay empty until microphone permission. */
16 +export async function listAudioInputs(): Promise<BrowserMicrophone[]> {
17 const devices = await navigator.mediaDevices.enumerateDevices();
18 return devices
19 .filter((device) => device.kind === "audioinput")
7 - .map(({ deviceId, label }, index) => ({
8 - deviceId,
9 - label: label || `Microphone ${index + 1}`,
10 - }));
20 + .map(({ deviceId, label }) => ({ deviceId, label }));
21 +}
22 +
23 +export async function listBrowserMicrophones(): Promise<BrowserMicrophone[]> {
24 + return (await listAudioInputs()).map(({ deviceId, label }, index) => ({
25 + deviceId,
26 + label: label || `Microphone ${index + 1}`,
27 + }));
28 +}
29 +
30 +/** Applies a change to the saved order. Skipped before hydration so defaults never replace a saved order. */
31 +export function updateMicrophoneOrder(
32 + change: (order: ReadonlyArray<MicrophoneEntry>) => ReadonlyArray<MicrophoneEntry>,
33 +): ReadonlyArray<MicrophoneEntry> {
34 + const known = microphoneOrder(getClientSettings());
35 + if (!isClientSettingsHydrated()) return known;
36 + const next = change(known);
37 + if (next !== known) void updateClientSettings({ voiceMicrophones: [...next] });
38 + return next;
39 +}
40 +
41 +/** Remembers newly seen microphones and returns the device id to record from, empty for System default. */
42 +export async function chooseRecordingMicrophone(): Promise<string> {
43 + // Rejects where the browser has no media devices, such as plain HTTP.
44 + const devices = await listAudioInputs().catch(() => []);
45 + return preferredMicrophoneId(
46 + updateMicrophoneOrder((order) => rememberMicrophones(order, devices)),
47 + devices,
48 + );
49 }
apps/web/src/speech/browserPcmCapture.browser.test.ts
+29 -5
@@ -1,21 +1,30 @@
1 import { expect, it, vi } from "vite-plus/test";
2 -import { createBrowserPcmCapture } from "./browserPcmCapture";
2 +import { createBrowserPcmCapture, openMicrophone } from "./browserPcmCapture";
3
4 -it("cancels an unanswered permission prompt and stops a late microphone stream", async () => {
4 +it("records from the chosen microphone, cancels an unanswered prompt, and stops a late stream", async () => {
5 let grantPermission!: (media: MediaStream) => void;
6 const permission = new Promise<MediaStream>((resolve) => {
7 grantPermission = resolve;
8 });
9 using microphone = vi.spyOn(navigator.mediaDevices, "getUserMedia").mockReturnValue(permission);
10 const capture = createBrowserPcmCapture(
11 - { getMicrophoneId: () => "", onLevel: vi.fn(), onDurationLimit: vi.fn(), onError: vi.fn() },
11 + {
12 + getMicrophoneId: async () => "blue-yeti",
13 + onLevel: vi.fn(),
14 + onDurationLimit: vi.fn(),
15 + onError: vi.fn(),
16 + },
17 async () => ({ locale: "en", transcribe: async () => "" }),
18 );
19 const abort = new AbortController();
20 await capture.transcriber.prepare({ signal: abort.signal });
21 const rejected = vi.fn();
22 const preparing = capture.recorder.prepareToRecordAsync(abort.signal).catch(rejected);
18 - expect(microphone).toHaveBeenCalledOnce();
23 + await vi.waitFor(() =>
24 + expect(microphone).toHaveBeenCalledExactlyOnceWith({
25 + audio: { deviceId: { exact: "blue-yeti" } },
26 + }),
27 + );
28 abort.abort();
29 const stop = vi.fn();
30 try {
@@ -24,5 +33,20 @@ it("cancels an unanswered permission prompt and stops a late microphone stream",
33 grantPermission({ getTracks: () => [{ stop }] } as unknown as MediaStream);
34 await preparing;
35 }
27 - expect(stop).toHaveBeenCalledOnce();
36 + await vi.waitFor(() => expect(stop).toHaveBeenCalledOnce());
37 +});
38 +
39 +it("falls back to the system default when the saved microphone is gone, and reports other failures", async () => {
40 + const media = {} as MediaStream;
41 + using microphone = vi
42 + .spyOn(navigator.mediaDevices, "getUserMedia")
43 + .mockRejectedValueOnce(new DOMException("gone", "OverconstrainedError"))
44 + .mockResolvedValueOnce(media);
45 + await expect(openMicrophone("blue-yeti")).resolves.toBe(media);
46 + expect(microphone).toHaveBeenLastCalledWith({ audio: true });
47 +
48 + const denied = new DOMException("denied", "NotAllowedError");
49 + microphone.mockRejectedValueOnce(denied);
50 + await expect(openMicrophone("blue-yeti")).rejects.toBe(denied);
51 + expect(microphone).toHaveBeenCalledTimes(3);
52 });
apps/web/src/speech/browserPcmCapture.ts
+18 -5
@@ -10,9 +10,23 @@ type PreparedBrowserTranscription = PreparedVoiceTranscription & {
10 readonly streaming?: VoiceStreamingSession & { readonly feed: (pcm: Float32Array) => void };
11 };
12
13 +/** Opens a microphone. A saved device the browser can no longer find falls back to System default. */
14 +export async function openMicrophone(microphoneId: string): Promise<MediaStream> {
15 + if (!microphoneId) return navigator.mediaDevices.getUserMedia({ audio: true });
16 + try {
17 + return await navigator.mediaDevices.getUserMedia({
18 + audio: { deviceId: { exact: microphoneId } },
19 + });
20 + } catch (error) {
21 + const name = typeof error === "object" && error !== null && "name" in error ? error.name : "";
22 + if (name !== "OverconstrainedError" && name !== "NotFoundError") throw error;
23 + return navigator.mediaDevices.getUserMedia({ audio: true });
24 + }
25 +}
26 +
27 export function createBrowserPcmCapture(
28 input: {
15 - readonly getMicrophoneId: () => string;
29 + readonly getMicrophoneId: () => string | Promise<string>;
30 readonly onLevel: (level: number) => void;
31 readonly onDurationLimit: () => void;
32 readonly onError: (message: string) => void;
@@ -59,12 +73,11 @@ export function createBrowserPcmCapture(
73 if (recordingUri) URL.revokeObjectURL(recordingUri);
74 recordingUri = null;
75 chunks = [];
62 - const microphoneId = input.getMicrophoneId();
76 + const microphoneId = await input.getMicrophoneId();
77 + captureSignal.throwIfAborted();
78 const captured = await new Promise<MediaStream>((resolve, reject) => {
79 const onAbort = () => reject(captureSignal.reason);
65 - const permission = navigator.mediaDevices.getUserMedia({
66 - audio: microphoneId ? { deviceId: { exact: microphoneId } } : true,
67 - });
80 + const permission = openMicrophone(microphoneId);
81 captureSignal.addEventListener("abort", onAbort, { once: true });
82 void permission.then(
83 (media) => {
apps/web/src/speech/browserVoiceInput.ts
+1 -1
@@ -27,7 +27,7 @@ export function createBrowserVoiceInputPlatform(input: {
27 readonly cloudModel?: CloudSpeechModelId | undefined;
28 readonly hasAudioConsent: () => boolean;
29 readonly getTranscriptionOptions: () => SpeechTranscriptionOptions;
30 - readonly getMicrophoneId: () => string;
30 + readonly getMicrophoneId: () => string | Promise<string>;
31 readonly onLevel: (level: number) => void;
32 readonly onDurationLimit: () => void;
33 readonly onText: (text: SpeechStreamText) => void;
apps/web/src/speech/microphoneOrder.ts new
+128
@@ -0,0 +1,128 @@
1 +import type { ClientSettings, VoiceMicrophoneEntry } from "@t3tools/contracts";
2 +
3 +import type { BrowserMicrophone } from "./browserMicrophones";
4 +
5 +export type MicrophoneEntry = VoiceMicrophoneEntry;
6 +
7 +const SYSTEM_DEFAULT: MicrophoneEntry = { kind: "default" };
8 +
9 +// Chromium lists the current default and Windows communications device under these ids too.
10 +const ALIAS_DEVICE_IDS = new Set(["", "default", "communications"]);
11 +
12 +/** True for ids that stand for the system default, including the id-less input shown before permission. */
13 +export function isSystemDefaultDeviceId(deviceId: string): boolean {
14 + return ALIAS_DEVICE_IDS.has(deviceId);
15 +}
16 +
17 +export function microphoneEntryKey(entry: MicrophoneEntry): string {
18 + return entry.kind === "default" ? "default" : `device:${entry.deviceId}`;
19 +}
20 +
21 +/** The saved order with exactly one System default row; an empty order starts from the legacy choice. */
22 +export function microphoneOrder(
23 + settings: Pick<ClientSettings, "voiceMicrophones" | "voiceMicrophone">,
24 +): ReadonlyArray<MicrophoneEntry> {
25 + if (settings.voiceMicrophones.length === 0) {
26 + return settings.voiceMicrophone
27 + ? [{ kind: "device", deviceId: settings.voiceMicrophone, label: "" }, SYSTEM_DEFAULT]
28 + : [SYSTEM_DEFAULT];
29 + }
30 + const keys = new Set<string>();
31 + const order = settings.voiceMicrophones.filter((entry) => {
32 + const key = microphoneEntryKey(entry);
33 + if (keys.has(key)) return false;
34 + keys.add(key);
35 + return true;
36 + });
37 + return keys.has("default") ? order : [...order, SYSTEM_DEFAULT];
38 +}
39 +
40 +/** Adds labelled devices seen for the first time directly below System default and refreshes labels. */
41 +export function rememberMicrophones(
42 + order: ReadonlyArray<MicrophoneEntry>,
43 + devices: ReadonlyArray<BrowserMicrophone>,
44 +): ReadonlyArray<MicrophoneEntry> {
45 + let next: MicrophoneEntry[] | null = null;
46 + let insertAt = order.findIndex((entry) => entry.kind === "default") + 1;
47 + for (const device of devices) {
48 + if (!device.label || ALIAS_DEVICE_IDS.has(device.deviceId)) continue;
49 + const current: ReadonlyArray<MicrophoneEntry> = next ?? order;
50 + const index = current.findIndex(
51 + (entry) => entry.kind === "device" && entry.deviceId === device.deviceId,
52 + );
53 + const existing = current[index];
54 + if (existing?.kind === "device" && existing.label === device.label) continue;
55 + next ??= [...order];
56 + const entry: MicrophoneEntry = {
57 + kind: "device",
58 + deviceId: device.deviceId,
59 + label: device.label,
60 + };
61 + if (index !== -1) {
62 + next[index] = entry;
63 + continue;
64 + }
65 + next.splice(insertAt, 0, entry);
66 + insertAt += 1;
67 + }
68 + return next ?? order;
69 +}
70 +
71 +/**
72 + * The first entry that is connected right now. System default always is. Before microphone
73 + * permission the browser hides device ids, so the first entry is used as it was saved.
74 + */
75 +export function preferredMicrophone(
76 + order: ReadonlyArray<MicrophoneEntry>,
77 + devices: ReadonlyArray<BrowserMicrophone>,
78 +): MicrophoneEntry {
79 + if (devices.every((device) => !device.deviceId)) return order[0] ?? SYSTEM_DEFAULT;
80 + return (
81 + order.find(
82 + (entry) =>
83 + entry.kind === "default" || devices.some((device) => device.deviceId === entry.deviceId),
84 + ) ?? SYSTEM_DEFAULT
85 + );
86 +}
87 +
88 +/** The `deviceId` to record from, or an empty string for System default. */
89 +export function preferredMicrophoneId(
90 + order: ReadonlyArray<MicrophoneEntry>,
91 + devices: ReadonlyArray<BrowserMicrophone>,
92 +): string {
93 + const entry = preferredMicrophone(order, devices);
94 + return entry.kind === "device" ? entry.deviceId : "";
95 +}
96 +
97 +export function moveMicrophone(
98 + order: ReadonlyArray<MicrophoneEntry>,
99 + key: string,
100 + steps: number,
101 +): ReadonlyArray<MicrophoneEntry> {
102 + const from = order.findIndex((entry) => microphoneEntryKey(entry) === key);
103 + const to = Math.max(0, Math.min(order.length - 1, from + steps));
104 + if (from === -1 || from === to) return order;
105 + const next = [...order];
106 + const [entry] = next.splice(from, 1);
107 + if (!entry) return order;
108 + next.splice(to, 0, entry);
109 + return next;
110 +}
111 +
112 +/** Moves an entry to the top, adding it when it is not listed yet. */
113 +export function preferMicrophone(
114 + order: ReadonlyArray<MicrophoneEntry>,
115 + entry: MicrophoneEntry,
116 +): ReadonlyArray<MicrophoneEntry> {
117 + const key = microphoneEntryKey(entry);
118 + const existing = order.find((candidate) => microphoneEntryKey(candidate) === key);
119 + return [existing ?? entry, ...order.filter((candidate) => microphoneEntryKey(candidate) !== key)];
120 +}
121 +
122 +/** Removes a named microphone. System default stays. */
123 +export function forgetMicrophone(
124 + order: ReadonlyArray<MicrophoneEntry>,
125 + key: string,
126 +): ReadonlyArray<MicrophoneEntry> {
127 + return key === "default" ? order : order.filter((entry) => microphoneEntryKey(entry) !== key);
128 +}
apps/web/src/speech/useEnvironmentSpeechInput.browser.test.tsx
+2
@@ -91,6 +91,8 @@ vi.mock("../state/environments", () => ({
91 }));
92 vi.mock("../hooks/useSettings", () => ({
93 updateClientSettings: vi.fn(),
94 + getClientSettings: () => ({ ...DEFAULT_CLIENT_SETTINGS, ...mocks.settings }),
95 + isClientSettingsHydrated: () => true,
96 useClientSettingsHydrated: () => true,
97 useClientSettings: (selector?: (settings: Record<string, unknown>) => unknown) => {
98 const settings = { ...DEFAULT_CLIENT_SETTINGS, ...mocks.settings };
apps/web/src/speech/useEnvironmentSpeechInput.ts
+2 -1
@@ -37,6 +37,7 @@ import {
37 } from "../hooks/useSettings";
38 import { useEnvironment } from "../state/environments";
39 import { useEnvironmentScope, usePreparedConnection } from "../state/session";
40 +import { chooseRecordingMicrophone } from "./browserMicrophones";
41 import { createBrowserVoiceInputPlatform } from "./browserVoiceInput";
42 import { preferredSpeechLanguages } from "./preferredSpeechLanguages";
43 import { runEnvironmentSpeech } from "./runEnvironmentSpeech";
@@ -296,7 +297,7 @@ export function useEnvironmentSpeechInput(input: HookInput) {
297 speechSilenceFiltering: settings.speechSilenceFiltering,
298 };
299 },
299 - getMicrophoneId: () => latestRef.current.clientSettings.voiceMicrophone,
300 + getMicrophoneId: chooseRecordingMicrophone,
301 onLevel: setLevel,
302 onDurationLimit: () => void controller.stop(),
303 onText: (text) => {
apps/web/src/speech/useTextFieldSpeech.tsx
+23 -22
@@ -75,35 +75,36 @@ export function TextFieldSpeechControls({
75 return (
76 <>
77 {speech.preview ? (
78 - <p role="status" className="w-full text-sm text-muted-foreground">
78 + <p
79 + role="status"
80 + className="order-first max-h-24 w-full overflow-y-auto rounded-lg border border-border/60 bg-muted/30 px-2.5 py-1.5 text-sm break-words whitespace-pre-wrap"
81 + >
82 {speech.preview.committed}
80 - <span className="opacity-70">{speech.preview.tentative}</span>
83 + <span className="text-muted-foreground">{speech.preview.tentative}</span>
84 </p>
85 ) : null}
83 - <div className="flex min-w-0 items-center gap-1">
84 - <ComposerSpeechRecordingPill
86 + <ComposerSpeechRecordingPill
87 + size="compact"
88 + state={speech.state}
89 + progress={null}
90 + level={speech.level}
91 + finishShortcutLabel={speech.shortcutLabel}
92 + cancelShortcutLabel="Esc"
93 + disabled={disabled}
94 + onStart={() => void speech.start()}
95 + onStop={() => void speech.stop()}
96 + onCancel={speech.cancel}
97 + onSkipPostProcessing={speech.skipPostProcessing}
98 + />
99 + {speech.available ? (
100 + <ComposerSpeechButton
101 state={speech.state}
86 - progress={null}
87 - level={speech.level}
88 - finishShortcutLabel={speech.shortcutLabel}
89 - cancelShortcutLabel="Esc"
102 + size="compact"
103 + shortcutLabel={speech.shortcutLabel}
104 disabled={disabled}
105 onStart={() => void speech.start()}
92 - onStop={() => void speech.stop()}
93 - onCancel={speech.cancel}
94 - onSkipPostProcessing={speech.skipPostProcessing}
106 />
96 - {speech.available ? (
97 - <ComposerSpeechButton
98 - state={speech.state}
99 - progress={null}
100 - shortcutLabel={speech.shortcutLabel}
101 - disabled={disabled}
102 - onStart={() => void speech.start()}
103 - onCancel={speech.cancel}
104 - />
105 - ) : null}
106 - </div>
107 + ) : null}
108 <VoiceInputSetup
109 environmentId={speech.transcriptionEnvironmentId}
110 open={speech.setup.open}
docs/user/voice-input.md
+8 -5
@@ -142,12 +142,15 @@ the model hears the cue as a dictionary hint.
142
143 ## Test and troubleshoot
144
145 -In **Settings → Voice**, **Test microphone** beside the microphone selector
146 -records up to ten seconds and plays it back in your browser;
147 -the audio stays on this device and is never transcribed, so silent playback
145 +In **Settings → Voice**, voice input uses the first connected microphone in the
146 +**Microphone** list. A microphone joins the list below System default the first
147 +time it is seen, so nothing changes until you move it up. If a preferred
148 +microphone is unplugged, the next connected one is used.
149 +
150 +**Test microphone** records from that microphone for up to ten seconds and
151 +plays it back in your browser; the audio stays on this device and is never transcribed, so silent playback
152 points at the microphone. **Test model** on the selected model shows a transcript using the
153 same environment, language, and dictionary as the composer, but skips transcript
154 cleanup even when it is on, so you see what the model heard. Browser recording
155 needs HTTPS or localhost and microphone permission; for a remote server over
152 -plain HTTP, use the desktop app. If the selected microphone is missing, choose
153 -another or the system default.
156 +plain HTTP, use the desktop app.
packages/contracts/src/settings.ts
+12 -1
@@ -370,6 +370,12 @@ export type ModelPreferences = {
370 export const VoiceShortcutMode = Schema.Literals(["auto", "hold", "toggle"]);
371 export type VoiceShortcutMode = typeof VoiceShortcutMode.Type;
372
373 +export const VoiceMicrophoneEntry = Schema.Union([
374 + Schema.Struct({ kind: Schema.Literal("default") }),
375 + Schema.Struct({ kind: Schema.Literal("device"), deviceId: Schema.String, label: Schema.String }),
376 +]);
377 +export type VoiceMicrophoneEntry = typeof VoiceMicrophoneEntry.Type;
378 +
379 /**
380 * Where this device sends microphone audio. A cloud grant names the forwarding environment and
381 * the provider; an OpenRouter grant also names the exact model, since OpenRouter has no default.
@@ -548,8 +554,12 @@ export const ClientSettingsSchema = Schema.Struct({
554 timestampFormat: TimestampFormat.pipe(
555 Schema.withDecodingDefault(Effect.succeed(DEFAULT_TIMESTAMP_FORMAT)),
556 ),
551 - // Device-local input device id. An empty string follows the operating system default.
557 + // Superseded by voiceMicrophones; read only to seed an empty order.
558 voiceMicrophone: Schema.String.pipe(Schema.withDecodingDefault(Effect.succeed(""))),
559 + // Device-local microphone order. Recording uses the first connected entry.
560 + voiceMicrophones: Schema.Array(VoiceMicrophoneEntry).pipe(
561 + Schema.withDecodingDefault(Effect.succeed([])),
562 + ),
563 voiceShortcutMode: VoiceShortcutMode.pipe(Schema.withDecodingDefault(Effect.succeed("auto"))),
564 speechLanguage: SpeechLanguage.pipe(Schema.withDecodingDefault(Effect.succeed("auto"))),
565 speechTranslateToEnglish: Schema.Boolean.pipe(Schema.withDecodingDefault(Effect.succeed(false))),
@@ -2050,6 +2060,7 @@ export const ClientSettingsPatch = Schema.Struct({
2060 sidebarThreadSortOrder: Schema.optionalKey(SidebarThreadSortOrder),
2061 timestampFormat: Schema.optionalKey(TimestampFormat),
2062 voiceMicrophone: Schema.optionalKey(Schema.String),
2063 + voiceMicrophones: Schema.optionalKey(Schema.Array(VoiceMicrophoneEntry)),
2064 voiceShortcutMode: Schema.optionalKey(VoiceShortcutMode),
2065 speechLanguage: Schema.optionalKey(SpeechLanguage),
2066 speechTranslateToEnglish: Schema.optionalKey(Schema.Boolean),