Files
openchamber/packages/web/server/lib/dictation/service.js
T
Bohdan Triapitsyn de1b85ac56 feat(voice): first-class voice input and local TTS across web, desktop, and mobile (#2018)
Complete rebuild of voice input on a server-authoritative streaming
architecture, replacing the legacy Web Speech / whole-blob / WASM engines
and the dead voice-agent layer (~4k lines removed).

Speech-to-text (dictation):
- Client streams 16 kHz mono PCM16 chunks over /api/dictation/ws with
  seq/ack ordering; buffered audio is retained and replayed on reconnect
- Server transcribes and streams live partial transcripts back;
  segments auto-commit every ~15s with silence suppression and adaptive
  finalization timeouts
- Local provider (default, zero config): sherpa-onnx models in a forked
  worker process — auto-download with progress, staged extraction with
  verification, corrupt-model auto-recovery, idle shutdown after 5 min
- Model catalog with settings picker (accuracy/speed ratings, sizes,
  download/delete): Parakeet TDT v2 (English) and v3 (25 European
  languages, auto-detected), Whisper base and tiny (multilingual, light)
- OpenAI-compatible provider for any Whisper endpoint
- Composer overlay with live transcript, volume meter, timer, and
  cancel / insert / insert-and-send actions; failed transcriptions keep
  their audio for retry or accepting the partial text as-is
- Configurable keyboard shortcut (default mod+alt+v) toggles dictation;
  Enter confirms and Escape cancels while recording
- Overlay is pixel-aligned with the composer (measured footer height,
  matching paddings/typography/gaps) — no layout shift when toggling

Text-to-speech:
- Local Kokoro provider (English, 11 voices) synthesized in the same
  worker via /api/dictation/tts/speak, managed by the shared model
  pipeline; sentence-pipelined playback keeps time-to-first-audio at
  ~1 sentence regardless of message length, and stop cancels in-flight
  synthesis
- Sanitizer keeps inline-code content (strips backticks only), reads
  interword slashes aloud, and removes only absolute file paths

Settings:
- Voice page unified: a single read-aloud toggle owns all playback
  options (the confusing "Enable Voice Mode" is gone); a new "Enable
  voice input" toggle (default on, persisted to settings.json) hides
  the composer mic entirely when disabled

Mobile and transport:
- iOS/Android microphone permissions added (dictation was previously
  impossible on mobile)
- Fixed Android WebSocket upgrades: the Capacitor WebView origin
  (https://localhost) was missing from the packaged-client allowlist,
  403-ing every WS connection — root cause of the old mobile SSE lock,
  which is now removed for all transports

Security and conventions:
- All HTTP routes sit behind the global /api auth gate; the WS upgrade
  explicitly validates the UI session and origin, with oc_url_token
  narrowly allowlisted and covered by tests; the dictation socket mints
  a fresh URL token before connecting
- Routes register before the generic OpenCode proxy; the client goes
  through runtimeFetch/getRuntimeUrlResolver, and runtime switches
  reset the dictation socket
- VS Code deliberately reports dictation as unavailable (no server
  process in that runtime)

CI: workflow Node bumped 20 -> 22 to match the repo engines and fix
better-sqlite3 installs broken by node-gyp@latest on Node 20.

New dependency: sherpa-onnx-node (prebuilt N-API; macOS/Linux x64+arm64,
Windows x64 — Windows-on-ARM falls back to the OpenAI-compatible provider)
2026-07-04 02:48:07 +03:00

303 lines
9.8 KiB
JavaScript

/**
* Dictation service: resolves STT providers, tracks local model download
* state, and exposes a readiness snapshot for the status route.
*
* Providers:
* - 'local' (default): sherpa-onnx Parakeet running in a worker process.
* Models auto-download in the background on first use.
* - 'openai-compatible': any OpenAI-compatible /v1/audio/transcriptions
* endpoint (faster-whisper, whisper.cpp, OpenAI).
*/
import { rm } from 'fs/promises';
import { DictationWorkerClient, WorkerBackedTranscriptionSession } from './local/worker-client.js';
import { OpenAICompatibleTranscriptionSession } from './openai-compatible-session.js';
import {
DEFAULT_LOCAL_STT_MODEL,
DEFAULT_LOCAL_TTS_MODEL,
LOCAL_STT_MODEL_CATALOG,
LOCAL_STT_MODEL_IDS,
LOCAL_TTS_MODEL_CATALOG,
LOCAL_TTS_MODEL_IDS,
getLocalSttModelDir,
isLocalModelId,
isLocalSttModelId,
isLocalTtsModelId,
} from './local/model-catalog.js';
import { ensureLocalSttModel, isLocalSttModelInstalled } from './local/model-downloader.js';
export function createDictationService({ modelsDir }) {
const workerClient = new DictationWorkerClient();
/** modelId -> 'downloading' | 'error' */
const downloadStates = new Map();
/** modelId -> last download error message */
const downloadErrors = new Map();
/** modelId -> in-flight ensure promise */
const downloadPromises = new Map();
/** modelId -> 0..100 download percent (null while size unknown) */
const downloadProgress = new Map();
const startModelDownload = (modelId) => {
const existing = downloadPromises.get(modelId);
if (existing) {
return existing;
}
downloadStates.set(modelId, 'downloading');
downloadErrors.delete(modelId);
downloadProgress.set(modelId, 0);
const promise = ensureLocalSttModel({
modelsDir,
modelId,
onProgress: (downloadedBytes, totalBytes) => {
downloadProgress.set(
modelId,
totalBytes ? Math.min(100, Math.round((downloadedBytes / totalBytes) * 100)) : null,
);
},
})
.then(() => {
downloadStates.delete(modelId);
downloadPromises.delete(modelId);
downloadProgress.delete(modelId);
})
.catch((error) => {
downloadStates.set(modelId, 'error');
downloadErrors.set(modelId, error?.message || String(error));
downloadPromises.delete(modelId);
downloadProgress.delete(modelId);
});
downloadPromises.set(modelId, promise);
return promise;
};
const resolveLocalModelId = (requested) => {
return isLocalSttModelId(requested) ? requested : DEFAULT_LOCAL_STT_MODEL;
};
/**
* Create a connected StreamingTranscriptionSession for one dictation.
* Returns { session } on success or { error, retryable, reasonCode } when
* the provider is not ready.
*
* @param {{ provider?: string, language?: string, localModel?: string,
* openaiCompatible?: { baseUrl?: string, model?: string, apiKey?: string } }} options
*/
const createSttSession = async (options = {}) => {
const provider = options.provider === 'openai-compatible' ? 'openai-compatible' : 'local';
if (provider === 'openai-compatible') {
const config = options.openaiCompatible || {};
const session = new OpenAICompatibleTranscriptionSession({
baseURL: config.baseUrl,
model: config.model,
apiKey: config.apiKey || undefined,
language: options.language || undefined,
});
try {
await session.connect();
} catch (error) {
return {
error: error?.message || String(error),
retryable: false,
reasonCode: 'stt_not_configured',
};
}
return { session };
}
const modelId = resolveLocalModelId(options.localModel);
const installed = await isLocalSttModelInstalled(modelsDir, modelId);
if (!installed) {
const state = downloadStates.get(modelId);
if (state === 'error') {
const message = downloadErrors.get(modelId) || 'Model download failed';
// Allow a retry on the next attempt.
downloadStates.delete(modelId);
return {
error: `Failed to download dictation model: ${message}`,
retryable: true,
reasonCode: 'model_download_failed',
};
}
void startModelDownload(modelId);
return {
error: 'Dictation model is downloading',
retryable: true,
reasonCode: 'model_download_in_progress',
};
}
const session = new WorkerBackedTranscriptionSession(workerClient, { modelsDir, modelId });
try {
await session.connect();
} catch (error) {
const message = error?.message || String(error);
// A model that passes the file-presence check but fails to load is
// corrupt on disk (e.g. truncated by an interrupted extraction). Remove
// it so the next attempt re-downloads instead of crashing forever.
if (/Load model|Protobuf parsing failed/i.test(message)) {
await rm(getLocalSttModelDir(modelsDir, modelId), { recursive: true, force: true })
.catch(() => undefined);
return {
error: 'Dictation model files were corrupt and have been removed; retry to re-download',
retryable: true,
reasonCode: 'model_corrupt',
};
}
return {
error: message,
retryable: true,
reasonCode: 'stt_unavailable',
};
}
return { session };
};
/**
* Readiness snapshot for the status route and UI gating.
* @param {{ provider?: string, localModel?: string }} [options]
*/
const getStatus = async (options = {}) => {
const provider = options.provider === 'openai-compatible' ? 'openai-compatible' : 'local';
const modelId = resolveLocalModelId(options.localModel);
const describeModel = async (id, catalog) => ({
id,
description: catalog[id].description,
installed: await isLocalSttModelInstalled(modelsDir, id),
downloading: downloadStates.get(id) === 'downloading',
downloadProgress: downloadProgress.get(id) ?? null,
downloadError: downloadErrors.get(id) || null,
});
const models = await Promise.all(
LOCAL_STT_MODEL_IDS.map((id) => describeModel(id, LOCAL_STT_MODEL_CATALOG)),
);
const ttsModels = await Promise.all(
LOCAL_TTS_MODEL_IDS.map((id) => describeModel(id, LOCAL_TTS_MODEL_CATALOG)),
);
if (provider === 'openai-compatible') {
return { provider, available: true, models, ttsModels };
}
const model = models.find((entry) => entry.id === modelId) || null;
if (model?.installed) {
return { provider, available: true, activeModel: modelId, models, ttsModels };
}
if (model?.downloading) {
return {
provider,
available: false,
reasonCode: 'model_download_in_progress',
activeModel: modelId,
models,
ttsModels,
};
}
if (model?.downloadError) {
return {
provider,
available: false,
reasonCode: 'model_download_failed',
error: model.downloadError,
activeModel: modelId,
models,
ttsModels,
};
}
return {
provider,
available: false,
reasonCode: 'models_missing',
activeModel: modelId,
models,
ttsModels,
};
};
/**
* Synthesize speech with the local TTS model. Returns WAV bytes, or a
* readiness error while the model is missing/downloading.
* @param {{ text: string, model?: string, speakerId?: number, speed?: number }} options
*/
const synthesizeSpeech = async ({ text, model, speakerId, speed }) => {
const modelId = isLocalTtsModelId(model) ? model : DEFAULT_LOCAL_TTS_MODEL;
const installed = await isLocalSttModelInstalled(modelsDir, modelId);
if (!installed) {
const state = downloadStates.get(modelId);
if (state === 'error') {
const message = downloadErrors.get(modelId) || 'Model download failed';
downloadStates.delete(modelId);
return {
error: `Failed to download TTS model: ${message}`,
retryable: true,
reasonCode: 'model_download_failed',
};
}
void startModelDownload(modelId);
return {
error: 'TTS model is downloading',
retryable: true,
reasonCode: 'model_download_in_progress',
};
}
const result = await workerClient.synthesizeSpeech({
modelsDir,
modelId,
text,
speakerId,
speed,
});
return { audio: result.audio, format: result.format };
};
/**
* Kick off a background download for a model (used by the status route's
* download action so Settings can pre-download models).
*/
const requestModelDownload = async (modelId) => {
if (!isLocalModelId(modelId)) {
return { ok: false, error: 'Unknown model id' };
}
if (await isLocalSttModelInstalled(modelsDir, modelId)) {
return { ok: true, installed: true };
}
void startModelDownload(modelId);
return { ok: true, installed: false };
};
/**
* Delete an installed model from disk. A model that is mid-download cannot
* be deleted. An engine already loaded in the worker keeps its in-memory
* copy until the worker's idle shutdown; the files are simply re-downloaded
* on the next use if the model is selected again.
*/
const deleteModel = async (modelId) => {
if (!isLocalModelId(modelId)) {
return { ok: false, error: 'Unknown model id' };
}
if (downloadStates.get(modelId) === 'downloading') {
return { ok: false, error: 'Model is downloading' };
}
await rm(getLocalSttModelDir(modelsDir, modelId), { recursive: true, force: true });
downloadErrors.delete(modelId);
return { ok: true };
};
const shutdown = () => {
workerClient.shutdown();
};
return {
createSttSession,
synthesizeSpeech,
getStatus,
requestModelDownload,
deleteModel,
shutdown,
};
}