Files
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

137 lines
4.0 KiB
JavaScript

/**
* Loader for the sherpa-onnx-node native addon.
*
* sherpa-onnx-node ships its native addon and shared libraries in a
* platform-specific package (e.g. sherpa-onnx-darwin-arm64). The shared
* libraries must be findable via the platform's dynamic-loader search path,
* so the loader prepends the platform package directory to LD_LIBRARY_PATH /
* DYLD_LIBRARY_PATH / PATH before requiring the addon.
*/
import { createRequire } from 'module';
import path from 'path';
import { existsSync } from 'fs';
const require = createRequire(import.meta.url);
let cached = null;
function sherpaPlatformPackageName(platform = process.platform, arch = process.arch) {
const normalizedPlatform = platform === 'win32' ? 'win' : platform;
return `sherpa-onnx-${normalizedPlatform}-${arch}`;
}
function sherpaLoaderEnvKey(platform = process.platform) {
if (platform === 'linux') {
return 'LD_LIBRARY_PATH';
}
if (platform === 'darwin') {
return 'DYLD_LIBRARY_PATH';
}
if (platform === 'win32') {
return 'PATH';
}
return null;
}
function prependEnvPath(existing, value) {
const parts = String(existing ?? '').split(path.delimiter).filter(Boolean);
if (parts.includes(value)) {
return parts.join(path.delimiter);
}
return [value, ...parts].join(path.delimiter);
}
/**
* Case-insensitive env key lookup: on Windows `{...process.env}` yields a
* plain object where PATH may be stored as `Path`. Using a hardcoded 'PATH'
* would create a duplicate key and break the child process PATH.
*/
function findEnvKey(env, key) {
const lower = key.toLowerCase();
for (const k of Object.keys(env)) {
if (k.toLowerCase() === lower) {
return k;
}
}
return key;
}
function resolveSherpaLibDir(platform = process.platform, arch = process.arch) {
const packageName = sherpaPlatformPackageName(platform, arch);
try {
const pkgJson = require.resolve(`${packageName}/package.json`);
// Electron packages node_modules inside app.asar, but native addons and
// their shared libraries are extracted to app.asar.unpacked. The dynamic
// loader (dlopen/DYLD/LD) cannot read from the asar archive, so point the
// search path at the unpacked copy.
const dir = path.dirname(pkgJson);
const unpacked = dir.replace(`app.asar${path.sep}`, `app.asar.unpacked${path.sep}`);
return existsSync(unpacked) ? unpacked : dir;
} catch {
return null;
}
}
/**
* Prepend the sherpa platform package dir to the loader search path env var.
* Mutates the provided env object.
* @param {NodeJS.ProcessEnv} env
*/
export function applySherpaLoaderEnv(env) {
const key = sherpaLoaderEnvKey();
const libDir = resolveSherpaLibDir();
if (!key || !libDir) {
return { key: null, libDir: null };
}
const actualKey = findEnvKey(env, key);
env[actualKey] = prependEnvPath(env[actualKey], libDir);
return { key, libDir };
}
/**
* Load the sherpa-onnx-node module, trying the upstream entry first and then
* the platform addon directly.
*/
export function loadSherpaOnnxNode() {
if (cached) {
return cached;
}
const attempts = [];
try {
cached = require('sherpa-onnx-node');
return cached;
} catch (error) {
attempts.push(`sherpa-onnx-node: ${error?.message || String(error)}`);
}
const libDir = resolveSherpaLibDir();
if (libDir) {
applySherpaLoaderEnv(process.env);
const addonPath = path.join(libDir, 'sherpa-onnx.node');
if (existsSync(addonPath)) {
try {
cached = require(addonPath);
return cached;
} catch (error) {
attempts.push(`${addonPath}: ${error?.message || String(error)}`);
}
} else {
attempts.push(`${addonPath}: file not found`);
}
} else {
attempts.push(`${sherpaPlatformPackageName()}: platform package not installed`);
}
throw new Error(
[
`Failed to load sherpa-onnx-node for ${process.platform}-${process.arch}.`,
`Node ${process.version} (ABI ${process.versions.modules}).`,
'Load attempts:',
...attempts.map((line) => `- ${line}`),
].join('\n'),
);
}