feat: implement Whisper worker for isolated audio transcription
This commit is contained in:
Илья Глазунов
parent
1b74968006
commit
684b61755c
161
src/index.js
161
src/index.js
@ -1,11 +1,43 @@
|
||||
if (require('electron-squirrel-startup')) {
|
||||
if (require("electron-squirrel-startup")) {
|
||||
process.exit(0);
|
||||
}
|
||||
|
||||
const { app, BrowserWindow, shell, ipcMain } = require('electron');
|
||||
const { createWindow, updateGlobalShortcuts } = require('./utils/window');
|
||||
const { setupGeminiIpcHandlers, stopMacOSAudioCapture, sendToRenderer } = require('./utils/gemini');
|
||||
const storage = require('./storage');
|
||||
// ── Global crash handlers to prevent silent process termination ──
|
||||
process.on("uncaughtException", (error) => {
|
||||
console.error("[FATAL] Uncaught exception:", error);
|
||||
try {
|
||||
const { sendToRenderer } = require("./utils/gemini");
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Fatal error: " + (error?.message || "unknown"),
|
||||
);
|
||||
} catch (_) {
|
||||
// sendToRenderer may not be available yet
|
||||
}
|
||||
});
|
||||
|
||||
process.on("unhandledRejection", (reason) => {
|
||||
console.error("[FATAL] Unhandled promise rejection:", reason);
|
||||
try {
|
||||
const { sendToRenderer } = require("./utils/gemini");
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Unhandled error: " +
|
||||
(reason instanceof Error ? reason.message : String(reason)),
|
||||
);
|
||||
} catch (_) {
|
||||
// sendToRenderer may not be available yet
|
||||
}
|
||||
});
|
||||
|
||||
const { app, BrowserWindow, shell, ipcMain } = require("electron");
|
||||
const { createWindow, updateGlobalShortcuts } = require("./utils/window");
|
||||
const {
|
||||
setupGeminiIpcHandlers,
|
||||
stopMacOSAudioCapture,
|
||||
sendToRenderer,
|
||||
} = require("./utils/gemini");
|
||||
const storage = require("./storage");
|
||||
|
||||
const geminiSessionRef = { current: null };
|
||||
let mainWindow = null;
|
||||
@ -20,9 +52,9 @@ app.whenReady().then(async () => {
|
||||
storage.initializeStorage();
|
||||
|
||||
// Trigger screen recording permission prompt on macOS if not already granted
|
||||
if (process.platform === 'darwin') {
|
||||
const { desktopCapturer } = require('electron');
|
||||
desktopCapturer.getSources({ types: ['screen'] }).catch(() => {});
|
||||
if (process.platform === "darwin") {
|
||||
const { desktopCapturer } = require("electron");
|
||||
desktopCapturer.getSources({ types: ["screen"] }).catch(() => {});
|
||||
}
|
||||
|
||||
createMainWindow();
|
||||
@ -31,18 +63,18 @@ app.whenReady().then(async () => {
|
||||
setupGeneralIpcHandlers();
|
||||
});
|
||||
|
||||
app.on('window-all-closed', () => {
|
||||
app.on("window-all-closed", () => {
|
||||
stopMacOSAudioCapture();
|
||||
if (process.platform !== 'darwin') {
|
||||
if (process.platform !== "darwin") {
|
||||
app.quit();
|
||||
}
|
||||
});
|
||||
|
||||
app.on('before-quit', () => {
|
||||
app.on("before-quit", () => {
|
||||
stopMacOSAudioCapture();
|
||||
});
|
||||
|
||||
app.on('activate', () => {
|
||||
app.on("activate", () => {
|
||||
if (BrowserWindow.getAllWindows().length === 0) {
|
||||
createMainWindow();
|
||||
}
|
||||
@ -50,250 +82,255 @@ app.on('activate', () => {
|
||||
|
||||
function setupStorageIpcHandlers() {
|
||||
// ============ CONFIG ============
|
||||
ipcMain.handle('storage:get-config', async () => {
|
||||
ipcMain.handle("storage:get-config", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getConfig() };
|
||||
} catch (error) {
|
||||
console.error('Error getting config:', error);
|
||||
console.error("Error getting config:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:set-config', async (event, config) => {
|
||||
ipcMain.handle("storage:set-config", async (event, config) => {
|
||||
try {
|
||||
storage.setConfig(config);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error setting config:', error);
|
||||
console.error("Error setting config:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:update-config', async (event, key, value) => {
|
||||
ipcMain.handle("storage:update-config", async (event, key, value) => {
|
||||
try {
|
||||
storage.updateConfig(key, value);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error updating config:', error);
|
||||
console.error("Error updating config:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// ============ CREDENTIALS ============
|
||||
ipcMain.handle('storage:get-credentials', async () => {
|
||||
ipcMain.handle("storage:get-credentials", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getCredentials() };
|
||||
} catch (error) {
|
||||
console.error('Error getting credentials:', error);
|
||||
console.error("Error getting credentials:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:set-credentials', async (event, credentials) => {
|
||||
ipcMain.handle("storage:set-credentials", async (event, credentials) => {
|
||||
try {
|
||||
storage.setCredentials(credentials);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error setting credentials:', error);
|
||||
console.error("Error setting credentials:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:get-api-key', async () => {
|
||||
ipcMain.handle("storage:get-api-key", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getApiKey() };
|
||||
} catch (error) {
|
||||
console.error('Error getting API key:', error);
|
||||
console.error("Error getting API key:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:set-api-key', async (event, apiKey) => {
|
||||
ipcMain.handle("storage:set-api-key", async (event, apiKey) => {
|
||||
try {
|
||||
storage.setApiKey(apiKey);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error setting API key:', error);
|
||||
console.error("Error setting API key:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:get-groq-api-key', async () => {
|
||||
ipcMain.handle("storage:get-groq-api-key", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getGroqApiKey() };
|
||||
} catch (error) {
|
||||
console.error('Error getting Groq API key:', error);
|
||||
console.error("Error getting Groq API key:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:set-groq-api-key', async (event, groqApiKey) => {
|
||||
ipcMain.handle("storage:set-groq-api-key", async (event, groqApiKey) => {
|
||||
try {
|
||||
storage.setGroqApiKey(groqApiKey);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error setting Groq API key:', error);
|
||||
console.error("Error setting Groq API key:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// ============ PREFERENCES ============
|
||||
ipcMain.handle('storage:get-preferences', async () => {
|
||||
ipcMain.handle("storage:get-preferences", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getPreferences() };
|
||||
} catch (error) {
|
||||
console.error('Error getting preferences:', error);
|
||||
console.error("Error getting preferences:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:set-preferences', async (event, preferences) => {
|
||||
ipcMain.handle("storage:set-preferences", async (event, preferences) => {
|
||||
try {
|
||||
storage.setPreferences(preferences);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error setting preferences:', error);
|
||||
console.error("Error setting preferences:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:update-preference', async (event, key, value) => {
|
||||
ipcMain.handle("storage:update-preference", async (event, key, value) => {
|
||||
try {
|
||||
storage.updatePreference(key, value);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error updating preference:', error);
|
||||
console.error("Error updating preference:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// ============ KEYBINDS ============
|
||||
ipcMain.handle('storage:get-keybinds', async () => {
|
||||
ipcMain.handle("storage:get-keybinds", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getKeybinds() };
|
||||
} catch (error) {
|
||||
console.error('Error getting keybinds:', error);
|
||||
console.error("Error getting keybinds:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:set-keybinds', async (event, keybinds) => {
|
||||
ipcMain.handle("storage:set-keybinds", async (event, keybinds) => {
|
||||
try {
|
||||
storage.setKeybinds(keybinds);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error setting keybinds:', error);
|
||||
console.error("Error setting keybinds:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// ============ HISTORY ============
|
||||
ipcMain.handle('storage:get-all-sessions', async () => {
|
||||
ipcMain.handle("storage:get-all-sessions", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getAllSessions() };
|
||||
} catch (error) {
|
||||
console.error('Error getting sessions:', error);
|
||||
console.error("Error getting sessions:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:get-session', async (event, sessionId) => {
|
||||
ipcMain.handle("storage:get-session", async (event, sessionId) => {
|
||||
try {
|
||||
return { success: true, data: storage.getSession(sessionId) };
|
||||
} catch (error) {
|
||||
console.error('Error getting session:', error);
|
||||
console.error("Error getting session:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:save-session', async (event, sessionId, data) => {
|
||||
ipcMain.handle("storage:save-session", async (event, sessionId, data) => {
|
||||
try {
|
||||
storage.saveSession(sessionId, data);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error saving session:', error);
|
||||
console.error("Error saving session:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:delete-session', async (event, sessionId) => {
|
||||
ipcMain.handle("storage:delete-session", async (event, sessionId) => {
|
||||
try {
|
||||
storage.deleteSession(sessionId);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error deleting session:', error);
|
||||
console.error("Error deleting session:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('storage:delete-all-sessions', async () => {
|
||||
ipcMain.handle("storage:delete-all-sessions", async () => {
|
||||
try {
|
||||
storage.deleteAllSessions();
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error deleting all sessions:', error);
|
||||
console.error("Error deleting all sessions:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// ============ LIMITS ============
|
||||
ipcMain.handle('storage:get-today-limits', async () => {
|
||||
ipcMain.handle("storage:get-today-limits", async () => {
|
||||
try {
|
||||
return { success: true, data: storage.getTodayLimits() };
|
||||
} catch (error) {
|
||||
console.error('Error getting today limits:', error);
|
||||
console.error("Error getting today limits:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
// ============ CLEAR ALL ============
|
||||
ipcMain.handle('storage:clear-all', async () => {
|
||||
ipcMain.handle("storage:clear-all", async () => {
|
||||
try {
|
||||
storage.clearAllData();
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error clearing all data:', error);
|
||||
console.error("Error clearing all data:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
function setupGeneralIpcHandlers() {
|
||||
ipcMain.handle('get-app-version', async () => {
|
||||
ipcMain.handle("get-app-version", async () => {
|
||||
return app.getVersion();
|
||||
});
|
||||
|
||||
ipcMain.handle('quit-application', async event => {
|
||||
ipcMain.handle("quit-application", async (event) => {
|
||||
try {
|
||||
stopMacOSAudioCapture();
|
||||
app.quit();
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error quitting application:', error);
|
||||
console.error("Error quitting application:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.handle('open-external', async (event, url) => {
|
||||
ipcMain.handle("open-external", async (event, url) => {
|
||||
try {
|
||||
await shell.openExternal(url);
|
||||
return { success: true };
|
||||
} catch (error) {
|
||||
console.error('Error opening external URL:', error);
|
||||
console.error("Error opening external URL:", error);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
});
|
||||
|
||||
ipcMain.on('update-keybinds', (event, newKeybinds) => {
|
||||
ipcMain.on("update-keybinds", (event, newKeybinds) => {
|
||||
if (mainWindow) {
|
||||
// Also save to storage
|
||||
storage.setKeybinds(newKeybinds);
|
||||
updateGlobalShortcuts(newKeybinds, mainWindow, sendToRenderer, geminiSessionRef);
|
||||
updateGlobalShortcuts(
|
||||
newKeybinds,
|
||||
mainWindow,
|
||||
sendToRenderer,
|
||||
geminiSessionRef,
|
||||
);
|
||||
}
|
||||
});
|
||||
|
||||
// Debug logging from renderer
|
||||
ipcMain.on('log-message', (event, msg) => {
|
||||
ipcMain.on("log-message", (event, msg) => {
|
||||
console.log(msg);
|
||||
});
|
||||
}
|
||||
|
||||
@ -1,17 +1,27 @@
|
||||
const { Ollama } = require('ollama');
|
||||
const { getSystemPrompt } = require('./prompts');
|
||||
const { sendToRenderer, initializeNewSession, saveConversationTurn } = require('./gemini');
|
||||
const { Ollama } = require("ollama");
|
||||
const { getSystemPrompt } = require("./prompts");
|
||||
const {
|
||||
sendToRenderer,
|
||||
initializeNewSession,
|
||||
saveConversationTurn,
|
||||
} = require("./gemini");
|
||||
const { fork } = require("child_process");
|
||||
const path = require("path");
|
||||
|
||||
// ── State ──
|
||||
|
||||
let ollamaClient = null;
|
||||
let ollamaModel = null;
|
||||
let whisperPipeline = null;
|
||||
let whisperWorker = null;
|
||||
let isWhisperLoading = false;
|
||||
let whisperReady = false;
|
||||
let localConversationHistory = [];
|
||||
let currentSystemPrompt = null;
|
||||
let isLocalActive = false;
|
||||
|
||||
// Pending transcription callback (one at a time)
|
||||
let pendingTranscribe = null;
|
||||
|
||||
// VAD state
|
||||
let isSpeaking = false;
|
||||
let speechBuffers = [];
|
||||
@ -20,13 +30,32 @@ let speechFrameCount = 0;
|
||||
|
||||
// VAD configuration
|
||||
const VAD_MODES = {
|
||||
NORMAL: { energyThreshold: 0.01, speechFramesRequired: 3, silenceFramesRequired: 30 },
|
||||
LOW_BITRATE: { energyThreshold: 0.008, speechFramesRequired: 4, silenceFramesRequired: 35 },
|
||||
AGGRESSIVE: { energyThreshold: 0.015, speechFramesRequired: 2, silenceFramesRequired: 20 },
|
||||
VERY_AGGRESSIVE: { energyThreshold: 0.02, speechFramesRequired: 2, silenceFramesRequired: 15 },
|
||||
NORMAL: {
|
||||
energyThreshold: 0.01,
|
||||
speechFramesRequired: 3,
|
||||
silenceFramesRequired: 30,
|
||||
},
|
||||
LOW_BITRATE: {
|
||||
energyThreshold: 0.008,
|
||||
speechFramesRequired: 4,
|
||||
silenceFramesRequired: 35,
|
||||
},
|
||||
AGGRESSIVE: {
|
||||
energyThreshold: 0.015,
|
||||
speechFramesRequired: 2,
|
||||
silenceFramesRequired: 20,
|
||||
},
|
||||
VERY_AGGRESSIVE: {
|
||||
energyThreshold: 0.02,
|
||||
speechFramesRequired: 2,
|
||||
silenceFramesRequired: 15,
|
||||
},
|
||||
};
|
||||
let vadConfig = VAD_MODES.VERY_AGGRESSIVE;
|
||||
|
||||
// Maximum speech buffer size: ~30 seconds at 16kHz, 16-bit mono
|
||||
const MAX_SPEECH_BUFFER_BYTES = 16000 * 2 * 30; // 960,000 bytes
|
||||
|
||||
// Audio resampling buffer
|
||||
let resampleRemainder = Buffer.alloc(0);
|
||||
|
||||
@ -47,15 +76,24 @@ function resample24kTo16k(inputBuffer) {
|
||||
const frac = srcPos - srcIndex;
|
||||
|
||||
const s0 = combined.readInt16LE(srcIndex * 2);
|
||||
const s1 = srcIndex + 1 < inputSamples ? combined.readInt16LE((srcIndex + 1) * 2) : s0;
|
||||
const s1 =
|
||||
srcIndex + 1 < inputSamples
|
||||
? combined.readInt16LE((srcIndex + 1) * 2)
|
||||
: s0;
|
||||
const interpolated = Math.round(s0 + frac * (s1 - s0));
|
||||
outputBuffer.writeInt16LE(Math.max(-32768, Math.min(32767, interpolated)), i * 2);
|
||||
outputBuffer.writeInt16LE(
|
||||
Math.max(-32768, Math.min(32767, interpolated)),
|
||||
i * 2,
|
||||
);
|
||||
}
|
||||
|
||||
// Store remainder for next call
|
||||
const consumedInputSamples = Math.ceil((outputSamples * 3) / 2);
|
||||
const remainderStart = consumedInputSamples * 2;
|
||||
resampleRemainder = remainderStart < combined.length ? combined.slice(remainderStart) : Buffer.alloc(0);
|
||||
resampleRemainder =
|
||||
remainderStart < combined.length
|
||||
? combined.slice(remainderStart)
|
||||
: Buffer.alloc(0);
|
||||
|
||||
return outputBuffer;
|
||||
}
|
||||
@ -84,8 +122,8 @@ function processVAD(pcm16kBuffer) {
|
||||
if (!isSpeaking && speechFrameCount >= vadConfig.speechFramesRequired) {
|
||||
isSpeaking = true;
|
||||
speechBuffers = [];
|
||||
console.log('[LocalAI] Speech started (RMS:', rms.toFixed(4), ')');
|
||||
sendToRenderer('update-status', 'Listening... (speech detected)');
|
||||
console.log("[LocalAI] Speech started (RMS:", rms.toFixed(4), ")");
|
||||
sendToRenderer("update-status", "Listening... (speech detected)");
|
||||
}
|
||||
} else {
|
||||
silenceFrameCount++;
|
||||
@ -93,13 +131,23 @@ function processVAD(pcm16kBuffer) {
|
||||
|
||||
if (isSpeaking && silenceFrameCount >= vadConfig.silenceFramesRequired) {
|
||||
isSpeaking = false;
|
||||
console.log('[LocalAI] Speech ended, accumulated', speechBuffers.length, 'chunks');
|
||||
sendToRenderer('update-status', 'Transcribing...');
|
||||
console.log(
|
||||
"[LocalAI] Speech ended, accumulated",
|
||||
speechBuffers.length,
|
||||
"chunks",
|
||||
);
|
||||
sendToRenderer("update-status", "Transcribing...");
|
||||
|
||||
// Trigger transcription with accumulated audio
|
||||
const audioData = Buffer.concat(speechBuffers);
|
||||
speechBuffers = [];
|
||||
handleSpeechEnd(audioData);
|
||||
handleSpeechEnd(audioData).catch((err) => {
|
||||
console.error("[LocalAI] handleSpeechEnd crashed:", err);
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Transcription error: " + (err?.message || "unknown"),
|
||||
);
|
||||
});
|
||||
return;
|
||||
}
|
||||
}
|
||||
@ -107,76 +155,252 @@ function processVAD(pcm16kBuffer) {
|
||||
// Accumulate audio during speech
|
||||
if (isSpeaking) {
|
||||
speechBuffers.push(Buffer.from(pcm16kBuffer));
|
||||
|
||||
// Cap buffer at ~30 seconds to prevent OOM and ONNX tensor overflow
|
||||
const totalBytes = speechBuffers.reduce((sum, b) => sum + b.length, 0);
|
||||
if (totalBytes >= MAX_SPEECH_BUFFER_BYTES) {
|
||||
isSpeaking = false;
|
||||
console.log(
|
||||
"[LocalAI] Speech buffer limit reached (" +
|
||||
totalBytes +
|
||||
" bytes), forcing transcription",
|
||||
);
|
||||
sendToRenderer("update-status", "Transcribing (max length reached)...");
|
||||
const audioData = Buffer.concat(speechBuffers);
|
||||
speechBuffers = [];
|
||||
silenceFrameCount = 0;
|
||||
speechFrameCount = 0;
|
||||
handleSpeechEnd(audioData).catch((err) => {
|
||||
console.error("[LocalAI] handleSpeechEnd crashed:", err);
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Transcription error: " + (err?.message || "unknown"),
|
||||
);
|
||||
});
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// ── Whisper Transcription ──
|
||||
// ── Whisper Worker (isolated child process) ──
|
||||
|
||||
function spawnWhisperWorker() {
|
||||
if (whisperWorker) return;
|
||||
|
||||
const workerPath = path.join(__dirname, "whisperWorker.js");
|
||||
console.log("[LocalAI] Spawning Whisper worker:", workerPath);
|
||||
|
||||
whisperWorker = fork(workerPath, [], {
|
||||
stdio: ["pipe", "pipe", "pipe", "ipc"],
|
||||
// ELECTRON_RUN_AS_NODE makes the Electron binary behave as plain Node.js,
|
||||
// which is required for child_process.fork() in packaged Electron apps.
|
||||
env: { ...process.env, ELECTRON_RUN_AS_NODE: "1" },
|
||||
});
|
||||
|
||||
whisperWorker.stdout.on("data", (data) => {
|
||||
console.log("[WhisperWorker stdout]", data.toString().trim());
|
||||
});
|
||||
whisperWorker.stderr.on("data", (data) => {
|
||||
console.error("[WhisperWorker stderr]", data.toString().trim());
|
||||
});
|
||||
|
||||
whisperWorker.on("message", (msg) => {
|
||||
switch (msg.type) {
|
||||
case "ready":
|
||||
console.log("[LocalAI] Whisper worker ready");
|
||||
break;
|
||||
case "load-result":
|
||||
handleWorkerLoadResult(msg);
|
||||
break;
|
||||
case "transcribe-result":
|
||||
handleWorkerTranscribeResult(msg);
|
||||
break;
|
||||
case "status":
|
||||
sendToRenderer("update-status", msg.message);
|
||||
break;
|
||||
}
|
||||
});
|
||||
|
||||
whisperWorker.on("exit", (code, signal) => {
|
||||
console.error(
|
||||
"[LocalAI] Whisper worker exited — code:",
|
||||
code,
|
||||
"signal:",
|
||||
signal,
|
||||
);
|
||||
whisperWorker = null;
|
||||
whisperReady = false;
|
||||
|
||||
// Reject any pending transcription
|
||||
if (pendingTranscribe) {
|
||||
pendingTranscribe.reject(
|
||||
new Error(
|
||||
"Whisper worker crashed (code: " + code + ", signal: " + signal + ")",
|
||||
),
|
||||
);
|
||||
pendingTranscribe = null;
|
||||
}
|
||||
|
||||
// If session is still active, inform the user and respawn
|
||||
if (isLocalActive) {
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Whisper crashed (signal: " +
|
||||
(signal || code) +
|
||||
"). Respawning worker...",
|
||||
);
|
||||
setTimeout(() => {
|
||||
if (isLocalActive) {
|
||||
respawnWhisperWorker();
|
||||
}
|
||||
}, 2000);
|
||||
}
|
||||
});
|
||||
|
||||
whisperWorker.on("error", (err) => {
|
||||
console.error("[LocalAI] Whisper worker error:", err);
|
||||
whisperWorker = null;
|
||||
whisperReady = false;
|
||||
});
|
||||
}
|
||||
|
||||
let pendingLoad = null;
|
||||
|
||||
function handleWorkerLoadResult(msg) {
|
||||
if (msg.success) {
|
||||
console.log("[LocalAI] Whisper model loaded successfully (in worker)");
|
||||
whisperReady = true;
|
||||
sendToRenderer("whisper-downloading", false);
|
||||
isWhisperLoading = false;
|
||||
if (pendingLoad) {
|
||||
pendingLoad.resolve(true);
|
||||
pendingLoad = null;
|
||||
}
|
||||
} else {
|
||||
console.error("[LocalAI] Whisper worker failed to load model:", msg.error);
|
||||
sendToRenderer("whisper-downloading", false);
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Failed to load Whisper model: " + msg.error,
|
||||
);
|
||||
isWhisperLoading = false;
|
||||
if (pendingLoad) {
|
||||
pendingLoad.resolve(false);
|
||||
pendingLoad = null;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
function handleWorkerTranscribeResult(msg) {
|
||||
if (!pendingTranscribe) return;
|
||||
if (msg.success) {
|
||||
console.log("[LocalAI] Transcription:", msg.text);
|
||||
pendingTranscribe.resolve(msg.text || null);
|
||||
} else {
|
||||
console.error("[LocalAI] Worker transcription error:", msg.error);
|
||||
pendingTranscribe.resolve(null);
|
||||
}
|
||||
pendingTranscribe = null;
|
||||
}
|
||||
|
||||
function respawnWhisperWorker() {
|
||||
killWhisperWorker();
|
||||
spawnWhisperWorker();
|
||||
const { app } = require("electron");
|
||||
const cacheDir = path.join(app.getPath("userData"), "whisper-models");
|
||||
const modelName =
|
||||
require("../storage").getPreferences().whisperModel ||
|
||||
"Xenova/whisper-small";
|
||||
sendToRenderer("whisper-downloading", true);
|
||||
isWhisperLoading = true;
|
||||
whisperWorker.send({ type: "load", modelName, cacheDir });
|
||||
}
|
||||
|
||||
function killWhisperWorker() {
|
||||
if (whisperWorker) {
|
||||
try {
|
||||
whisperWorker.removeAllListeners();
|
||||
whisperWorker.kill();
|
||||
} catch (_) {
|
||||
// Already dead
|
||||
}
|
||||
whisperWorker = null;
|
||||
whisperReady = false;
|
||||
}
|
||||
}
|
||||
|
||||
async function loadWhisperPipeline(modelName) {
|
||||
if (whisperPipeline) return whisperPipeline;
|
||||
if (whisperReady) return true;
|
||||
if (isWhisperLoading) return null;
|
||||
|
||||
isWhisperLoading = true;
|
||||
console.log('[LocalAI] Loading Whisper model:', modelName);
|
||||
sendToRenderer('whisper-downloading', true);
|
||||
sendToRenderer('update-status', 'Loading Whisper model (first time may take a while)...');
|
||||
console.log("[LocalAI] Loading Whisper model via worker:", modelName);
|
||||
sendToRenderer("whisper-downloading", true);
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Loading Whisper model (first time may take a while)...",
|
||||
);
|
||||
|
||||
try {
|
||||
// Dynamic import for ESM module
|
||||
const { pipeline, env } = await import('@huggingface/transformers');
|
||||
// Cache models outside the asar archive so ONNX runtime can load them
|
||||
const { app } = require('electron');
|
||||
const path = require('path');
|
||||
env.cacheDir = path.join(app.getPath('userData'), 'whisper-models');
|
||||
whisperPipeline = await pipeline('automatic-speech-recognition', modelName, {
|
||||
dtype: 'q8',
|
||||
device: 'auto',
|
||||
spawnWhisperWorker();
|
||||
|
||||
const { app } = require("electron");
|
||||
const cacheDir = path.join(app.getPath("userData"), "whisper-models");
|
||||
|
||||
return new Promise((resolve) => {
|
||||
pendingLoad = { resolve };
|
||||
whisperWorker.send({ type: "load", modelName, cacheDir });
|
||||
});
|
||||
console.log('[LocalAI] Whisper model loaded successfully');
|
||||
sendToRenderer('whisper-downloading', false);
|
||||
isWhisperLoading = false;
|
||||
return whisperPipeline;
|
||||
} catch (error) {
|
||||
console.error('[LocalAI] Failed to load Whisper model:', error);
|
||||
sendToRenderer('whisper-downloading', false);
|
||||
sendToRenderer('update-status', 'Failed to load Whisper model: ' + error.message);
|
||||
isWhisperLoading = false;
|
||||
return null;
|
||||
}
|
||||
}
|
||||
|
||||
function pcm16ToFloat32(pcm16Buffer) {
|
||||
const samples = pcm16Buffer.length / 2;
|
||||
const float32 = new Float32Array(samples);
|
||||
for (let i = 0; i < samples; i++) {
|
||||
float32[i] = pcm16Buffer.readInt16LE(i * 2) / 32768;
|
||||
}
|
||||
return float32;
|
||||
}
|
||||
|
||||
async function transcribeAudio(pcm16kBuffer) {
|
||||
if (!whisperPipeline) {
|
||||
console.error('[LocalAI] Whisper pipeline not loaded');
|
||||
if (!whisperReady || !whisperWorker) {
|
||||
console.error("[LocalAI] Whisper worker not ready");
|
||||
return null;
|
||||
}
|
||||
|
||||
if (!pcm16kBuffer || pcm16kBuffer.length < 2) {
|
||||
console.error("[LocalAI] Invalid audio buffer:", pcm16kBuffer?.length);
|
||||
return null;
|
||||
}
|
||||
|
||||
console.log(
|
||||
"[LocalAI] Starting transcription, audio length:",
|
||||
pcm16kBuffer.length,
|
||||
"bytes",
|
||||
);
|
||||
|
||||
// Send audio to worker as base64 (IPC serialization)
|
||||
const audioBase64 = pcm16kBuffer.toString("base64");
|
||||
|
||||
return new Promise((resolve, reject) => {
|
||||
// Timeout: if worker takes > 60s, assume it's stuck
|
||||
const timeout = setTimeout(() => {
|
||||
console.error("[LocalAI] Transcription timed out after 60s");
|
||||
if (pendingTranscribe) {
|
||||
pendingTranscribe = null;
|
||||
resolve(null);
|
||||
}
|
||||
}, 60000);
|
||||
|
||||
pendingTranscribe = {
|
||||
resolve: (val) => {
|
||||
clearTimeout(timeout);
|
||||
resolve(val);
|
||||
},
|
||||
reject: (err) => {
|
||||
clearTimeout(timeout);
|
||||
reject(err);
|
||||
},
|
||||
};
|
||||
|
||||
try {
|
||||
const float32Audio = pcm16ToFloat32(pcm16kBuffer);
|
||||
|
||||
// Whisper expects audio at 16kHz which is what we have
|
||||
const result = await whisperPipeline(float32Audio, {
|
||||
sampling_rate: 16000,
|
||||
language: 'en',
|
||||
task: 'transcribe',
|
||||
});
|
||||
|
||||
const text = result.text?.trim();
|
||||
console.log('[LocalAI] Transcription:', text);
|
||||
return text;
|
||||
} catch (error) {
|
||||
console.error('[LocalAI] Transcription error:', error);
|
||||
return null;
|
||||
whisperWorker.send({ type: "transcribe", audioBase64 });
|
||||
} catch (err) {
|
||||
clearTimeout(timeout);
|
||||
pendingTranscribe = null;
|
||||
console.error("[LocalAI] Failed to send to worker:", err);
|
||||
resolve(null);
|
||||
}
|
||||
});
|
||||
}
|
||||
|
||||
// ── Speech End Handler ──
|
||||
@ -186,35 +410,52 @@ async function handleSpeechEnd(audioData) {
|
||||
|
||||
// Minimum audio length check (~0.5 seconds at 16kHz, 16-bit)
|
||||
if (audioData.length < 16000) {
|
||||
console.log('[LocalAI] Audio too short, skipping');
|
||||
sendToRenderer('update-status', 'Listening...');
|
||||
console.log("[LocalAI] Audio too short, skipping");
|
||||
sendToRenderer("update-status", "Listening...");
|
||||
return;
|
||||
}
|
||||
|
||||
console.log("[LocalAI] Processing audio:", audioData.length, "bytes");
|
||||
|
||||
try {
|
||||
const transcription = await transcribeAudio(audioData);
|
||||
|
||||
if (!transcription || transcription.trim() === '' || transcription.trim().length < 2) {
|
||||
console.log('[LocalAI] Empty transcription, skipping');
|
||||
sendToRenderer('update-status', 'Listening...');
|
||||
if (
|
||||
!transcription ||
|
||||
transcription.trim() === "" ||
|
||||
transcription.trim().length < 2
|
||||
) {
|
||||
console.log("[LocalAI] Empty transcription, skipping");
|
||||
sendToRenderer("update-status", "Listening...");
|
||||
return;
|
||||
}
|
||||
|
||||
sendToRenderer('update-status', 'Generating response...');
|
||||
sendToRenderer("update-status", "Generating response...");
|
||||
await sendToOllama(transcription);
|
||||
} catch (error) {
|
||||
console.error("[LocalAI] handleSpeechEnd error:", error);
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Error: " + (error?.message || "transcription failed"),
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
// ── Ollama Chat ──
|
||||
|
||||
async function sendToOllama(transcription) {
|
||||
if (!ollamaClient || !ollamaModel) {
|
||||
console.error('[LocalAI] Ollama not configured');
|
||||
console.error("[LocalAI] Ollama not configured");
|
||||
return;
|
||||
}
|
||||
|
||||
console.log('[LocalAI] Sending to Ollama:', transcription.substring(0, 100) + '...');
|
||||
console.log(
|
||||
"[LocalAI] Sending to Ollama:",
|
||||
transcription.substring(0, 100) + "...",
|
||||
);
|
||||
|
||||
localConversationHistory.push({
|
||||
role: 'user',
|
||||
role: "user",
|
||||
content: transcription.trim(),
|
||||
});
|
||||
|
||||
@ -225,7 +466,10 @@ async function sendToOllama(transcription) {
|
||||
|
||||
try {
|
||||
const messages = [
|
||||
{ role: 'system', content: currentSystemPrompt || 'You are a helpful assistant.' },
|
||||
{
|
||||
role: "system",
|
||||
content: currentSystemPrompt || "You are a helpful assistant.",
|
||||
},
|
||||
...localConversationHistory,
|
||||
];
|
||||
|
||||
@ -235,41 +479,52 @@ async function sendToOllama(transcription) {
|
||||
stream: true,
|
||||
});
|
||||
|
||||
let fullText = '';
|
||||
let fullText = "";
|
||||
let isFirst = true;
|
||||
|
||||
for await (const part of response) {
|
||||
const token = part.message?.content || '';
|
||||
const token = part.message?.content || "";
|
||||
if (token) {
|
||||
fullText += token;
|
||||
sendToRenderer(isFirst ? 'new-response' : 'update-response', fullText);
|
||||
sendToRenderer(isFirst ? "new-response" : "update-response", fullText);
|
||||
isFirst = false;
|
||||
}
|
||||
}
|
||||
|
||||
if (fullText.trim()) {
|
||||
localConversationHistory.push({
|
||||
role: 'assistant',
|
||||
role: "assistant",
|
||||
content: fullText.trim(),
|
||||
});
|
||||
|
||||
saveConversationTurn(transcription, fullText);
|
||||
}
|
||||
|
||||
console.log('[LocalAI] Ollama response completed');
|
||||
sendToRenderer('update-status', 'Listening...');
|
||||
console.log("[LocalAI] Ollama response completed");
|
||||
sendToRenderer("update-status", "Listening...");
|
||||
} catch (error) {
|
||||
console.error('[LocalAI] Ollama error:', error);
|
||||
sendToRenderer('update-status', 'Ollama error: ' + error.message);
|
||||
console.error("[LocalAI] Ollama error:", error);
|
||||
sendToRenderer("update-status", "Ollama error: " + error.message);
|
||||
}
|
||||
}
|
||||
|
||||
// ── Public API ──
|
||||
|
||||
async function initializeLocalSession(ollamaHost, model, whisperModel, profile, customPrompt) {
|
||||
console.log('[LocalAI] Initializing local session:', { ollamaHost, model, whisperModel, profile });
|
||||
async function initializeLocalSession(
|
||||
ollamaHost,
|
||||
model,
|
||||
whisperModel,
|
||||
profile,
|
||||
customPrompt,
|
||||
) {
|
||||
console.log("[LocalAI] Initializing local session:", {
|
||||
ollamaHost,
|
||||
model,
|
||||
whisperModel,
|
||||
profile,
|
||||
});
|
||||
|
||||
sendToRenderer('session-initializing', true);
|
||||
sendToRenderer("session-initializing", true);
|
||||
|
||||
try {
|
||||
// Setup system prompt
|
||||
@ -282,18 +537,26 @@ async function initializeLocalSession(ollamaHost, model, whisperModel, profile,
|
||||
// Test Ollama connection
|
||||
try {
|
||||
await ollamaClient.list();
|
||||
console.log('[LocalAI] Ollama connection verified');
|
||||
console.log("[LocalAI] Ollama connection verified");
|
||||
} catch (error) {
|
||||
console.error('[LocalAI] Cannot connect to Ollama at', ollamaHost, ':', error.message);
|
||||
sendToRenderer('session-initializing', false);
|
||||
sendToRenderer('update-status', 'Cannot connect to Ollama at ' + ollamaHost);
|
||||
console.error(
|
||||
"[LocalAI] Cannot connect to Ollama at",
|
||||
ollamaHost,
|
||||
":",
|
||||
error.message,
|
||||
);
|
||||
sendToRenderer("session-initializing", false);
|
||||
sendToRenderer(
|
||||
"update-status",
|
||||
"Cannot connect to Ollama at " + ollamaHost,
|
||||
);
|
||||
return false;
|
||||
}
|
||||
|
||||
// Load Whisper model
|
||||
const pipeline = await loadWhisperPipeline(whisperModel);
|
||||
if (!pipeline) {
|
||||
sendToRenderer('session-initializing', false);
|
||||
sendToRenderer("session-initializing", false);
|
||||
return false;
|
||||
}
|
||||
|
||||
@ -309,15 +572,15 @@ async function initializeLocalSession(ollamaHost, model, whisperModel, profile,
|
||||
initializeNewSession(profile, customPrompt);
|
||||
|
||||
isLocalActive = true;
|
||||
sendToRenderer('session-initializing', false);
|
||||
sendToRenderer('update-status', 'Local AI ready - Listening...');
|
||||
sendToRenderer("session-initializing", false);
|
||||
sendToRenderer("update-status", "Local AI ready - Listening...");
|
||||
|
||||
console.log('[LocalAI] Session initialized successfully');
|
||||
console.log("[LocalAI] Session initialized successfully");
|
||||
return true;
|
||||
} catch (error) {
|
||||
console.error('[LocalAI] Initialization error:', error);
|
||||
sendToRenderer('session-initializing', false);
|
||||
sendToRenderer('update-status', 'Local AI error: ' + error.message);
|
||||
console.error("[LocalAI] Initialization error:", error);
|
||||
sendToRenderer("session-initializing", false);
|
||||
sendToRenderer("update-status", "Local AI error: " + error.message);
|
||||
return false;
|
||||
}
|
||||
}
|
||||
@ -333,7 +596,7 @@ function processLocalAudio(monoChunk24k) {
|
||||
}
|
||||
|
||||
function closeLocalSession() {
|
||||
console.log('[LocalAI] Closing local session');
|
||||
console.log("[LocalAI] Closing local session");
|
||||
isLocalActive = false;
|
||||
isSpeaking = false;
|
||||
speechBuffers = [];
|
||||
@ -344,7 +607,8 @@ function closeLocalSession() {
|
||||
ollamaClient = null;
|
||||
ollamaModel = null;
|
||||
currentSystemPrompt = null;
|
||||
// Note: whisperPipeline is kept loaded to avoid reloading on next session
|
||||
// Note: whisperWorker is kept alive to avoid reloading model on next session
|
||||
// To fully clean up, call killWhisperWorker()
|
||||
}
|
||||
|
||||
function isLocalSessionActive() {
|
||||
@ -355,7 +619,7 @@ function isLocalSessionActive() {
|
||||
|
||||
async function sendLocalText(text) {
|
||||
if (!isLocalActive || !ollamaClient) {
|
||||
return { success: false, error: 'No active local session' };
|
||||
return { success: false, error: "No active local session" };
|
||||
}
|
||||
|
||||
try {
|
||||
@ -368,28 +632,31 @@ async function sendLocalText(text) {
|
||||
|
||||
async function sendLocalImage(base64Data, prompt) {
|
||||
if (!isLocalActive || !ollamaClient) {
|
||||
return { success: false, error: 'No active local session' };
|
||||
return { success: false, error: "No active local session" };
|
||||
}
|
||||
|
||||
try {
|
||||
console.log('[LocalAI] Sending image to Ollama');
|
||||
sendToRenderer('update-status', 'Analyzing image...');
|
||||
console.log("[LocalAI] Sending image to Ollama");
|
||||
sendToRenderer("update-status", "Analyzing image...");
|
||||
|
||||
const userMessage = {
|
||||
role: 'user',
|
||||
role: "user",
|
||||
content: prompt,
|
||||
images: [base64Data],
|
||||
};
|
||||
|
||||
// Store text-only version in history
|
||||
localConversationHistory.push({ role: 'user', content: prompt });
|
||||
localConversationHistory.push({ role: "user", content: prompt });
|
||||
|
||||
if (localConversationHistory.length > 20) {
|
||||
localConversationHistory = localConversationHistory.slice(-20);
|
||||
}
|
||||
|
||||
const messages = [
|
||||
{ role: 'system', content: currentSystemPrompt || 'You are a helpful assistant.' },
|
||||
{
|
||||
role: "system",
|
||||
content: currentSystemPrompt || "You are a helpful assistant.",
|
||||
},
|
||||
...localConversationHistory.slice(0, -1),
|
||||
userMessage,
|
||||
];
|
||||
@ -400,29 +667,32 @@ async function sendLocalImage(base64Data, prompt) {
|
||||
stream: true,
|
||||
});
|
||||
|
||||
let fullText = '';
|
||||
let fullText = "";
|
||||
let isFirst = true;
|
||||
|
||||
for await (const part of response) {
|
||||
const token = part.message?.content || '';
|
||||
const token = part.message?.content || "";
|
||||
if (token) {
|
||||
fullText += token;
|
||||
sendToRenderer(isFirst ? 'new-response' : 'update-response', fullText);
|
||||
sendToRenderer(isFirst ? "new-response" : "update-response", fullText);
|
||||
isFirst = false;
|
||||
}
|
||||
}
|
||||
|
||||
if (fullText.trim()) {
|
||||
localConversationHistory.push({ role: 'assistant', content: fullText.trim() });
|
||||
localConversationHistory.push({
|
||||
role: "assistant",
|
||||
content: fullText.trim(),
|
||||
});
|
||||
saveConversationTurn(prompt, fullText);
|
||||
}
|
||||
|
||||
console.log('[LocalAI] Image response completed');
|
||||
sendToRenderer('update-status', 'Listening...');
|
||||
console.log("[LocalAI] Image response completed");
|
||||
sendToRenderer("update-status", "Listening...");
|
||||
return { success: true, text: fullText, model: ollamaModel };
|
||||
} catch (error) {
|
||||
console.error('[LocalAI] Image error:', error);
|
||||
sendToRenderer('update-status', 'Ollama error: ' + error.message);
|
||||
console.error("[LocalAI] Image error:", error);
|
||||
sendToRenderer("update-status", "Ollama error: " + error.message);
|
||||
return { success: false, error: error.message };
|
||||
}
|
||||
}
|
||||
|
||||
150
src/utils/whisperWorker.js
Normal file
150
src/utils/whisperWorker.js
Normal file
@ -0,0 +1,150 @@
|
||||
/**
|
||||
* Whisper Worker — runs ONNX Runtime in an isolated child process.
|
||||
*
|
||||
* The main Electron process forks this file and communicates via IPC messages.
|
||||
* If ONNX Runtime crashes (SIGSEGV/SIGABRT inside the native Metal or CPU
|
||||
* execution provider), only this worker dies — the main process survives and
|
||||
* can respawn the worker automatically.
|
||||
*
|
||||
* Protocol (parent ↔ worker):
|
||||
* parent → worker:
|
||||
* { type: 'load', modelName, cacheDir }
|
||||
* { type: 'transcribe', audioBase64 } // PCM 16-bit 16kHz as base64
|
||||
* { type: 'shutdown' }
|
||||
*
|
||||
* worker → parent:
|
||||
* { type: 'load-result', success, error? }
|
||||
* { type: 'transcribe-result', success, text?, error? }
|
||||
* { type: 'status', message }
|
||||
* { type: 'ready' }
|
||||
*/
|
||||
|
||||
let whisperPipeline = null;
|
||||
|
||||
function pcm16ToFloat32(pcm16Buffer) {
|
||||
if (!pcm16Buffer || pcm16Buffer.length === 0) {
|
||||
return new Float32Array(0);
|
||||
}
|
||||
const alignedLength =
|
||||
pcm16Buffer.length % 2 === 0 ? pcm16Buffer.length : pcm16Buffer.length - 1;
|
||||
const samples = alignedLength / 2;
|
||||
const float32 = new Float32Array(samples);
|
||||
for (let i = 0; i < samples; i++) {
|
||||
float32[i] = pcm16Buffer.readInt16LE(i * 2) / 32768;
|
||||
}
|
||||
return float32;
|
||||
}
|
||||
|
||||
async function loadModel(modelName, cacheDir) {
|
||||
if (whisperPipeline) {
|
||||
send({ type: "load-result", success: true });
|
||||
return;
|
||||
}
|
||||
|
||||
try {
|
||||
send({
|
||||
type: "status",
|
||||
message: "Loading Whisper model (first time may take a while)...",
|
||||
});
|
||||
const { pipeline, env } = await import("@huggingface/transformers");
|
||||
env.cacheDir = cacheDir;
|
||||
whisperPipeline = await pipeline(
|
||||
"automatic-speech-recognition",
|
||||
modelName,
|
||||
{
|
||||
dtype: "q8",
|
||||
device: "cpu",
|
||||
},
|
||||
);
|
||||
send({ type: "load-result", success: true });
|
||||
} catch (error) {
|
||||
send({ type: "load-result", success: false, error: error.message });
|
||||
}
|
||||
}
|
||||
|
||||
async function transcribe(audioBase64) {
|
||||
if (!whisperPipeline) {
|
||||
send({
|
||||
type: "transcribe-result",
|
||||
success: false,
|
||||
error: "Whisper pipeline not loaded",
|
||||
});
|
||||
return;
|
||||
}
|
||||
|
||||
try {
|
||||
const pcm16Buffer = Buffer.from(audioBase64, "base64");
|
||||
|
||||
if (pcm16Buffer.length < 2) {
|
||||
send({
|
||||
type: "transcribe-result",
|
||||
success: false,
|
||||
error: "Audio buffer too small",
|
||||
});
|
||||
return;
|
||||
}
|
||||
|
||||
// Cap at ~30 seconds (16kHz, 16-bit mono)
|
||||
const maxBytes = 16000 * 2 * 30;
|
||||
const audioData =
|
||||
pcm16Buffer.length > maxBytes
|
||||
? pcm16Buffer.slice(0, maxBytes)
|
||||
: pcm16Buffer;
|
||||
|
||||
const float32Audio = pcm16ToFloat32(audioData);
|
||||
if (float32Audio.length === 0) {
|
||||
send({
|
||||
type: "transcribe-result",
|
||||
success: false,
|
||||
error: "Empty audio after conversion",
|
||||
});
|
||||
return;
|
||||
}
|
||||
|
||||
const result = await whisperPipeline(float32Audio, {
|
||||
sampling_rate: 16000,
|
||||
language: "en",
|
||||
task: "transcribe",
|
||||
});
|
||||
|
||||
const text = result.text?.trim() || "";
|
||||
send({ type: "transcribe-result", success: true, text });
|
||||
} catch (error) {
|
||||
send({
|
||||
type: "transcribe-result",
|
||||
success: false,
|
||||
error: error.message || String(error),
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
function send(msg) {
|
||||
try {
|
||||
if (process.send) {
|
||||
process.send(msg);
|
||||
}
|
||||
} catch (_) {
|
||||
// Parent may have disconnected
|
||||
}
|
||||
}
|
||||
|
||||
process.on("message", (msg) => {
|
||||
switch (msg.type) {
|
||||
case "load":
|
||||
loadModel(msg.modelName, msg.cacheDir).catch((err) => {
|
||||
send({ type: "load-result", success: false, error: err.message });
|
||||
});
|
||||
break;
|
||||
case "transcribe":
|
||||
transcribe(msg.audioBase64).catch((err) => {
|
||||
send({ type: "transcribe-result", success: false, error: err.message });
|
||||
});
|
||||
break;
|
||||
case "shutdown":
|
||||
process.exit(0);
|
||||
break;
|
||||
}
|
||||
});
|
||||
|
||||
// Signal readiness to parent
|
||||
send({ type: "ready" });
|
||||
Loading…
x
Reference in New Issue
Block a user