Fix voice: Web Speech API for STT, WAV for TTS

- STT: Replace MediaRecorder+Whisper with browser Web Speech API (ko-KR) Whisper base model hallucinated English for Korean speech; Chrome's built-in Google STT is far more accurate for Korean - TTS: Skip ffmpeg OGG conversion, return WAV directly from Piper Avoids OGG/Opus codec compatibility issues in browsers Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-05-19 18:15:41 +09:00
parent de18ddc9f6
commit cd48a8ebe0
2 changed files with 61 additions and 66 deletions
@@ -8742,15 +8742,33 @@ app.post('/api/voice/stt', async (req: express.Request, res: express.Response) =
  req.on('error', (e: any) => res.status(500).json({ success: false, error: String(e?.message || e) }));
 });

-// ── Voice: TTS (Piper) ────────────────────────────────────────────────────
+// ── Voice: TTS (Piper → WAV) ─────────────────────────────────────────────
 app.post('/api/voice/tts', async (req: express.Request, res: express.Response) => {
  const { text } = req.body || {};
  if (!text || typeof text !== 'string') { res.status(400).json({ success: false, error: 'text required' }); return; }
-  if (!isTTSAvailable()) { res.status(503).json({ success: false, error: 'TTS not configured' }); return; }
+  const cfg = (getConfig().getConfig() as any)?.voice;
+  if (!cfg?.tts) { res.status(503).json({ success: false, error: 'TTS not configured' }); return; }
+  const piperPath = cfg.tts.piperPath || 'piper';
+  const modelPath = cfg.tts.modelPath || '';
+  const configPath = cfg.tts.configPath || '';
+  if (!modelPath) { res.status(503).json({ success: false, error: 'TTS model path not configured' }); return; }
  try {
-    const audio = await synthesizeSpeech(text.slice(0, 2000));
-    res.set('Content-Type', 'audio/ogg');
-    res.send(audio);
+    const os = await import('os');
+    const wavPath = path.join(os.tmpdir(), `tts_${Date.now()}.wav`);
+    const piperArgs = ['--model', modelPath, '--output_file', wavPath];
+    if (configPath) piperArgs.push('--config', configPath);
+    const { spawn: spawnProc } = await import('child_process');
+    await new Promise<void>((resolve, reject) => {
+      const piper = spawnProc(piperPath, piperArgs);
+      piper.stdin.write(text.slice(0, 2000), 'utf8');
+      piper.stdin.end();
+      piper.on('close', (code: number | null) => code === 0 ? resolve() : reject(new Error(`Piper exited ${code}`)));
+      piper.on('error', reject);
+    });
+    const wav = fs.readFileSync(wavPath);
+    try { fs.unlinkSync(wavPath); } catch {}
+    res.set('Content-Type', 'audio/wav');
+    res.send(wav);
  } catch (e: any) {
    res.status(500).json({ success: false, error: String(e?.message || e) });
  }
@@ -8832,11 +8832,9 @@ function initLuckyModalInteraction() {
  });
 }

-// ── Voice: STT (마이크 입력) ─────────────────────────────────────────────
-var _mediaRecorder = null;
-var _audioChunks = [];
+// ── Voice: STT (Web Speech API — Chrome 내장 Google STT) ────────────────
+var _speechRecog = null;
 var _isRecording = false;
-var _micMimeType = '';

 function toggleMicRecording() {
  if (_isRecording) stopMicRecording();
@@ -8844,66 +8842,45 @@ function toggleMicRecording() {
 }

 function startMicRecording() {
-  navigator.mediaDevices.getUserMedia({ audio: true, video: false })
-    .then(function(stream) {
-      _audioChunks = [];
-      _micMimeType = MediaRecorder.isTypeSupported('audio/webm;codecs=opus') ? 'audio/webm;codecs=opus'
-                   : MediaRecorder.isTypeSupported('audio/webm') ? 'audio/webm'
-                   : MediaRecorder.isTypeSupported('audio/ogg;codecs=opus') ? 'audio/ogg;codecs=opus'
-                   : '';
-      var opts = _micMimeType ? { mimeType: _micMimeType } : {};
-      _mediaRecorder = new MediaRecorder(stream, opts);
-      _mediaRecorder.ondataavailable = function(e) { if (e.data.size > 0) _audioChunks.push(e.data); };
-      _mediaRecorder.onstop = function() {
-        stream.getTracks().forEach(function(t) { t.stop(); });
-        var blob = new Blob(_audioChunks, { type: _micMimeType || 'audio/webm' });
-        sendToSTT(blob);
-      };
-      _mediaRecorder.start();
-      _isRecording = true;
-      var btn = document.getElementById('mic-btn');
-      if (btn) { btn.textContent = '⏹'; btn.title = '녹음 중지 (클릭)'; btn.classList.add('recording'); }
-    })
-    .catch(function(e) { alert('마이크 접근 실패: ' + e.message); });
+  var SpeechRecog = window.SpeechRecognition || window.webkitSpeechRecognition;
+  if (!SpeechRecog) {
+    alert('이 브라우저는 음성 인식을 지원하지 않습니다.\nChrome을 사용해주세요.');
+    return;
+  }
+  _speechRecog = new SpeechRecog();
+  _speechRecog.lang = 'ko-KR';
+  _speechRecog.interimResults = false;
+  _speechRecog.maxAlternatives = 1;
+  _speechRecog.continuous = false;
+
+  var btn = document.getElementById('mic-btn');
+
+  _speechRecog.onstart = function() {
+    _isRecording = true;
+    if (btn) { btn.textContent = '⏹'; btn.title = '말하는 중… (클릭 시 중지)'; btn.classList.add('recording'); }
+  };
+  _speechRecog.onresult = function(e) {
+    var transcript = e.results[0][0].transcript;
+    var input = document.getElementById('chat-input');
+    if (input && transcript) {
+      input.value = (input.value ? input.value + ' ' : '') + transcript;
+      input.focus();
+      input.dispatchEvent(new Event('input'));
+    }
+  };
+  _speechRecog.onerror = function(e) {
+    if (e.error !== 'no-speech') alert('음성 인식 오류: ' + e.error);
+  };
+  _speechRecog.onend = function() {
+    _isRecording = false;
+    _speechRecog = null;
+    if (btn) { btn.textContent = '🎤'; btn.title = '음성 입력'; btn.classList.remove('recording'); }
+  };
+  _speechRecog.start();
 }

 function stopMicRecording() {
-  if (_mediaRecorder && _isRecording) {
-    _mediaRecorder.stop();
-    _isRecording = false;
-    var btn = document.getElementById('mic-btn');
-    if (btn) { btn.textContent = '⌛'; btn.title = '인식 중…'; btn.classList.remove('recording'); }
-  }
-}
-
-function sendToSTT(blob) {
-  var ct = _micMimeType || 'audio/webm';
-  fetch('/api/voice/stt', {
-    method: 'POST',
-    headers: Object.assign({ 'Content-Type': ct }, authHeaders()),
-    credentials: 'include',
-    body: blob,
-  })
-  .then(function(r) { return r.json(); })
-  .then(function(data) {
-    var btn = document.getElementById('mic-btn');
-    if (btn) { btn.textContent = '🎤'; btn.title = '음성 입력 (클릭: 시작/중지)'; }
-    if (data.success && data.text) {
-      var input = document.getElementById('chat-input');
-      if (input) {
-        input.value = (input.value ? input.value + ' ' : '') + data.text;
-        input.focus();
-        input.dispatchEvent(new Event('input'));
-      }
-    } else if (!data.success) {
-      alert('음성 인식 실패: ' + (data.error || '알 수 없는 오류'));
-    }
-  })
-  .catch(function(e) {
-    var btn = document.getElementById('mic-btn');
-    if (btn) { btn.textContent = '🎤'; btn.title = '음성 입력 (클릭: 시작/중지)'; }
-    alert('STT 오류: ' + e.message);
-  });
+  if (_speechRecog) { _speechRecog.stop(); }
 }

 // ── Voice: TTS (음성 출력) ───────────────────────────────────────────────