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, 04:45 UTC
| 1 | import { cloudSpeechLanguages, effectiveSpeechLanguage } from "@t3tools/shared/speech"; |
| 2 | import { |
| 3 | requireCloudSpeechKey, |
| 4 | transcribeCloudPcm, |
| 5 | getEnvironmentSpeechStatus, |
| 6 | getEnvironmentSpeechStreamUrl, |
| 7 | openSpeechStream, |
| 8 | throwIfVoiceTranscriptionAborted, |
| 9 | transcribeEnvironmentPcm, |
| 10 | type VoiceInputErrorCode, |
| 11 | VoiceTranscriptionError, |
| 12 | type VoiceTranscriptionOptions, |
| 13 | } from "@t3tools/client-runtime/voice-input"; |
| 14 | import type { PreparedConnection } from "@t3tools/client-runtime/connection"; |
| 15 | import type { |
| 16 | SpeechTranscriptionOptions, |
| 17 | SpeechStreamText, |
| 18 | EnvironmentSpeechStatus, |
| 19 | } from "@t3tools/contracts"; |
| 20 | import type { CloudSpeechModelId, CloudSpeechProvider } from "@t3tools/contracts"; |
| 21 | import * as Schema from "effect/Schema"; |
| 22 | |
| 23 | import { readAppleSpeechLocales, resolveAppleSpeechLocale } from "./appleSpeech"; |
| 24 | import { createBrowserPcmCapture } from "./browserPcmCapture"; |
| 25 | import { runEnvironmentSpeech } from "./runEnvironmentSpeech"; |
| 26 | import { msg } from "gt-react"; |
| 27 | import { m } from "../i18n/messages"; |
| 28 | |
| 29 | const isVoiceTranscriptionError = Schema.is(VoiceTranscriptionError); |
| 30 | |
| 31 | export function createBrowserVoiceInputPlatform(input: { |
| 32 | readonly prepared: PreparedConnection; |
| 33 | readonly cloudProvider?: CloudSpeechProvider | undefined; |
| 34 | /** The exact OpenRouter model the grant names; required with that provider. */ |
| 35 | readonly cloudModel?: CloudSpeechModelId | undefined; |
| 36 | readonly hasAudioConsent: () => boolean; |
| 37 | readonly getTranscriptionOptions: () => SpeechTranscriptionOptions; |
| 38 | readonly getStatusListener?: () => ((status: EnvironmentSpeechStatus) => void) | undefined; |
| 39 | readonly getStreamingEnabled: () => boolean; |
| 40 | readonly getMicrophoneId: () => string | Promise<string>; |
| 41 | readonly onLevel: (level: number) => void; |
| 42 | readonly onDurationLimit: () => void; |
| 43 | readonly onText: (text: SpeechStreamText) => void; |
| 44 | readonly onError: (message: string, code?: VoiceInputErrorCode) => void; |
| 45 | }) { |
| 46 | return createBrowserPcmCapture(input, async (options, canStream) => { |
| 47 | const onStatus = input.getStatusListener?.(); |
| 48 | const recordingOptions = input.getTranscriptionOptions(); |
| 49 | const streamingEnabled = input.getStreamingEnabled(); |
| 50 | const transcribeRecording = async ( |
| 51 | uri: string, |
| 52 | { signal: transcriptionSignal }: VoiceTranscriptionOptions, |
| 53 | ) => { |
| 54 | try { |
| 55 | const response = await fetch(uri, { signal: transcriptionSignal }); |
| 56 | const pcm = new Uint8Array(await response.arrayBuffer()); |
| 57 | const result = await runEnvironmentSpeech( |
| 58 | input.cloudProvider |
| 59 | ? transcribeCloudPcm( |
| 60 | input.prepared, |
| 61 | input.cloudProvider, |
| 62 | pcm, |
| 63 | recordingOptions, |
| 64 | input.hasAudioConsent, |
| 65 | input.cloudModel, |
| 66 | ) |
| 67 | : transcribeEnvironmentPcm( |
| 68 | input.prepared, |
| 69 | pcm, |
| 70 | recordingOptions, |
| 71 | input.hasAudioConsent, |
| 72 | ), |
| 73 | { signal: transcriptionSignal }, |
| 74 | ); |
| 75 | throwIfVoiceTranscriptionAborted(transcriptionSignal); |
| 76 | return result.text; |
| 77 | } catch (cause) { |
| 78 | throwIfVoiceTranscriptionAborted(transcriptionSignal); |
| 79 | if (isVoiceTranscriptionError(cause)) throw cause; |
| 80 | throw new VoiceTranscriptionError({ |
| 81 | code: "transcription-failed", |
| 82 | message: m(msg("Voice transcription failed. Check the destination in Settings → Voice.")), |
| 83 | cause, |
| 84 | }); |
| 85 | } |
| 86 | }; |
| 87 | if (input.cloudProvider) { |
| 88 | if (input.cloudProvider === "openrouter" && !input.cloudModel) |
| 89 | throw new VoiceTranscriptionError({ |
| 90 | code: "unavailable", |
| 91 | message: m(msg("Choose an OpenRouter model in Settings → Voice.")), |
| 92 | }); |
| 93 | await runEnvironmentSpeech( |
| 94 | requireCloudSpeechKey(input.prepared, input.cloudProvider), |
| 95 | options, |
| 96 | ); |
| 97 | throwIfVoiceTranscriptionAborted(options.signal); |
| 98 | return { |
| 99 | locale: effectiveSpeechLanguage(cloudSpeechLanguages, recordingOptions.speechLanguage), |
| 100 | transcribe: transcribeRecording, |
| 101 | }; |
| 102 | } |
| 103 | const status = await runEnvironmentSpeech(getEnvironmentSpeechStatus(input.prepared), options); |
| 104 | throwIfVoiceTranscriptionAborted(options.signal); |
| 105 | onStatus?.(status); |
| 106 | throwIfVoiceTranscriptionAborted(options.signal); |
| 107 | if (!status.supported) |
| 108 | throw new VoiceTranscriptionError({ code: "unavailable", message: status.reason }); |
| 109 | if (status.state === "missing-model") |
| 110 | throw new VoiceTranscriptionError({ |
| 111 | code: "unavailable", |
| 112 | message: m(msg("Could not start voice input")), |
| 113 | }); |
| 114 | const language = status.languages |
| 115 | ? effectiveSpeechLanguage( |
| 116 | { |
| 117 | languages: status.languages, |
| 118 | supportsLanguageDetection: status.supportsLanguageDetection ?? false, |
| 119 | }, |
| 120 | recordingOptions.speechLanguage, |
| 121 | ) |
| 122 | : "auto"; |
| 123 | const nativeStreaming = status.supportsStreaming && streamingEnabled; |
| 124 | const batchLocale = |
| 125 | status.supportsTranslation && recordingOptions.speechTranslateToEnglish ? "en" : language; |
| 126 | const batch = { locale: batchLocale, transcribe: transcribeRecording }; |
| 127 | if ( |
| 128 | !canStream() || |
| 129 | (status.supportsStreaming ? !nativeStreaming : !status.supportsBufferedStreaming) |
| 130 | ) { |
| 131 | return batch; |
| 132 | } |
| 133 | const url = await runEnvironmentSpeech(getEnvironmentSpeechStreamUrl(input.prepared), options); |
| 134 | if (!canStream()) return batch; |
| 135 | return { |
| 136 | locale: nativeStreaming ? language : batchLocale, |
| 137 | transcribe: transcribeRecording, |
| 138 | streaming: openSpeechStream({ |
| 139 | url, |
| 140 | canSend: input.hasAudioConsent, |
| 141 | options: recordingOptions, |
| 142 | signal: options.signal, |
| 143 | onText: input.onText, |
| 144 | onError: (error) => input.onError(error.message), |
| 145 | }), |
| 146 | }; |
| 147 | }); |
| 148 | } |
| 149 | |
| 150 | async function waitForAppleSpeech<T>(request: Promise<T>, signal: AbortSignal): Promise<T> { |
| 151 | const aborted = Promise.withResolvers<never>(); |
| 152 | const onAbort = () => aborted.reject(signal.reason); |
| 153 | signal.addEventListener("abort", onAbort, { once: true }); |
| 154 | if (signal.aborted) onAbort(); |
| 155 | try { |
| 156 | return await Promise.race([request, aborted.promise]); |
| 157 | } finally { |
| 158 | signal.removeEventListener("abort", onAbort); |
| 159 | } |
| 160 | } |
| 161 | |
| 162 | /** Records like the environment path, then transcribes in the desktop main process; audio stays on this Mac. */ |
| 163 | export function createAppleSpeechVoiceInputPlatform(input: { |
| 164 | readonly hasAudioConsent: () => boolean; |
| 165 | readonly getTranscriptionOptions: () => SpeechTranscriptionOptions; |
| 166 | readonly getMicrophoneId: () => string | Promise<string>; |
| 167 | readonly onLevel: (level: number) => void; |
| 168 | readonly onDurationLimit: () => void; |
| 169 | readonly onError: (message: string, code?: VoiceInputErrorCode) => void; |
| 170 | }) { |
| 171 | return createBrowserPcmCapture(input, async ({ signal }) => { |
| 172 | const appleSpeech = window.desktopBridge?.appleSpeech; |
| 173 | const { speechLanguage } = input.getTranscriptionOptions(); |
| 174 | // The packaged app pins `navigator.language` to en-US, so "auto" asks the OS. |
| 175 | const system = window.desktopBridge?.getPreferredSystemLanguages?.() ?? []; |
| 176 | if (!appleSpeech) throw new Error("Apple Speech is not available."); |
| 177 | signal.throwIfAborted(); |
| 178 | const prepared = await waitForAppleSpeech( |
| 179 | appleSpeech.prepare( |
| 180 | resolveAppleSpeechLocale(readAppleSpeechLocales(), speechLanguage, system) ?? |
| 181 | navigator.language, |
| 182 | ), |
| 183 | signal, |
| 184 | ); |
| 185 | throwIfVoiceTranscriptionAborted(signal); |
| 186 | if (!prepared.ok) { |
| 187 | if (prepared.code === "failed") throw new Error("Apple Speech could not prepare."); |
| 188 | throw new VoiceTranscriptionError({ |
| 189 | code: prepared.code, |
| 190 | message: |
| 191 | prepared.code === "unsupported-locale" |
| 192 | ? m(msg("Apple Speech does not support this language.")) |
| 193 | : m(msg("Apple Speech is not available on this Mac.")), |
| 194 | }); |
| 195 | } |
| 196 | const locale = prepared.value; |
| 197 | return { |
| 198 | locale, |
| 199 | transcribe: async (uri, { signal: transcriptionSignal }) => { |
| 200 | const response = await fetch(uri, { signal: transcriptionSignal }); |
| 201 | const pcm = new Uint8Array(await response.arrayBuffer()); |
| 202 | throwIfVoiceTranscriptionAborted(transcriptionSignal); |
| 203 | if (!input.hasAudioConsent()) |
| 204 | throw new VoiceTranscriptionError({ |
| 205 | code: "cancelled", |
| 206 | message: m(msg("Voice transcription was cancelled.")), |
| 207 | }); |
| 208 | const result = await waitForAppleSpeech( |
| 209 | appleSpeech.transcribe({ locale, pcm }), |
| 210 | transcriptionSignal, |
| 211 | ); |
| 212 | throwIfVoiceTranscriptionAborted(transcriptionSignal); |
| 213 | if (!result.ok) throw new Error("Apple Speech could not transcribe."); |
| 214 | return result.value; |
| 215 | }, |
| 216 | }; |
| 217 | }); |
| 218 | } |