agentgui 1.0.168 → 1.0.169

package/lib/speech.js

@@ -13,6 +13,21 @@ const SPEAKER_EMBEDDINGS_PATH = path.join(DATA_DIR, 'speaker_embeddings.bin');
 const SAMPLE_RATE_TTS = 16000;
 const SAMPLE_RATE_STT = 16000;
 const MIN_WAV_SIZE = 44;
+const DATASET_API = 'https://datasets-server.huggingface.co/rows?dataset=Matthijs%2Fcmu-arctic-xvectors&config=default&split=validation';
+const SAMPLES_TO_AVERAGE = 10;
+
+const VOICE_CATALOG = [
+  { id: 'default', name: 'Default', gender: 'male', accent: 'US' },
+  { id: 'bdl', name: 'BDL', gender: 'male', accent: 'US' },
+  { id: 'slt', name: 'SLT', gender: 'female', accent: 'US' },
+  { id: 'clb', name: 'CLB', gender: 'female', accent: 'US' },
+  { id: 'rms', name: 'RMS', gender: 'male', accent: 'US' },
+  { id: 'awb', name: 'AWB', gender: 'male', accent: 'Scottish' },
+  { id: 'jmk', name: 'JMK', gender: 'male', accent: 'Canadian' },
+  { id: 'ksp', name: 'KSP', gender: 'male', accent: 'Indian' },
+];
+
+const SPEAKER_OFFSETS = { awb: 0, bdl: 1200, clb: 2300, jmk: 3500, ksp: 4700, rms: 5900, slt: 7100 };
 
 let transformersModule = null;
 let sttPipeline = null;
@@ -20,6 +35,7 @@ let ttsPipeline = null;
 let speakerEmbeddings = null;
 let sttLoading = false;
 let ttsLoading = false;
+const voiceEmbeddingsCache = new Map();
 
 const TTS_CACHE_MAX = 100;
 const ttsCache = new Map();
@@ -52,6 +68,43 @@ async function ensureSpeakerEmbeddings() {
   return speakerEmbeddings;
 }
 
+async function loadVoiceEmbedding(voiceId) {
+  if (!voiceId || voiceId === 'default') return ensureSpeakerEmbeddings();
+  if (voiceEmbeddingsCache.has(voiceId)) return voiceEmbeddingsCache.get(voiceId);
+  const binPath = path.join(DATA_DIR, `speaker_${voiceId}.bin`);
+  if (fs.existsSync(binPath)) {
+    const buf = fs.readFileSync(binPath);
+    const emb = new Float32Array(new Uint8Array(buf).buffer);
+    voiceEmbeddingsCache.set(voiceId, emb);
+    return emb;
+  }
+  const offset = SPEAKER_OFFSETS[voiceId];
+  if (offset === undefined) return ensureSpeakerEmbeddings();
+  const url = `${DATASET_API}&offset=${offset}&length=${SAMPLES_TO_AVERAGE}`;
+  const resp = await fetch(url);
+  if (!resp.ok) throw new Error('Failed to fetch voice embeddings for ' + voiceId);
+  const data = await resp.json();
+  const avg = new Float32Array(512);
+  let count = 0;
+  for (const item of data.rows) {
+    const match = item.row.filename.match(/cmu_us_(\w+)_arctic/);
+    if (match && match[1] === voiceId) {
+      for (let i = 0; i < 512; i++) avg[i] += item.row.xvector[i];
+      count++;
+    }
+  }
+  if (count === 0) return ensureSpeakerEmbeddings();
+  for (let i = 0; i < 512; i++) avg[i] /= count;
+  if (!fs.existsSync(DATA_DIR)) fs.mkdirSync(DATA_DIR, { recursive: true });
+  fs.writeFileSync(binPath, Buffer.from(avg.buffer));
+  voiceEmbeddingsCache.set(voiceId, avg);
+  return avg;
+}
+
+function getVoices() {
+  return VOICE_CATALOG;
+}
+
 async function getSTT() {
   if (sttPipeline) return sttPipeline;
   if (sttLoading) {
@@ -233,36 +286,38 @@ function cachePut(key, buf) {
   ttsCache.set(key, buf);
 }
 
-async function synthesize(text) {
-  const cached = ttsCache.get(text);
+async function synthesize(text, voiceId) {
+  const cacheKey = (voiceId || 'default') + ':' + text;
+  const cached = ttsCache.get(cacheKey);
   if (cached) {
-    ttsCache.delete(text);
-    ttsCache.set(text, cached);
+    ttsCache.delete(cacheKey);
+    ttsCache.set(cacheKey, cached);
     return cached;
   }
   const tts = await getTTS();
-  const embeddings = await ensureSpeakerEmbeddings();
+  const embeddings = await loadVoiceEmbedding(voiceId);
   const result = await tts(text, { speaker_embeddings: embeddings });
   const wav = encodeWav(result.audio, result.sampling_rate || SAMPLE_RATE_TTS);
-  cachePut(text, wav);
+  cachePut(cacheKey, wav);
   return wav;
 }
 
-async function* synthesizeStream(text) {
+async function* synthesizeStream(text, voiceId) {
   const sentences = splitSentences(text);
   const tts = await getTTS();
-  const embeddings = await ensureSpeakerEmbeddings();
+  const embeddings = await loadVoiceEmbedding(voiceId);
   for (const sentence of sentences) {
-    const cached = ttsCache.get(sentence);
+    const cacheKey = (voiceId || 'default') + ':' + sentence;
+    const cached = ttsCache.get(cacheKey);
     if (cached) {
-      ttsCache.delete(sentence);
-      ttsCache.set(sentence, cached);
+      ttsCache.delete(cacheKey);
+      ttsCache.set(cacheKey, cached);
       yield cached;
       continue;
     }
     const result = await tts(sentence, { speaker_embeddings: embeddings });
     const wav = encodeWav(result.audio, result.sampling_rate || SAMPLE_RATE_TTS);
-    cachePut(sentence, wav);
+    cachePut(cacheKey, wav);
     yield wav;
   }
 }
@@ -276,4 +331,4 @@ function getStatus() {
   };
 }
 
-export { transcribe, synthesize, synthesizeStream, getSTT, getTTS, getStatus };
+export { transcribe, synthesize, synthesizeStream, getSTT, getTTS, getStatus, getVoices };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentgui",
-  "version": "1.0.168",
+  "version": "1.0.169",
   "description": "Multi-agent ACP client with real-time communication",
   "type": "module",
   "main": "server.js",

package/server.js

@@ -535,16 +535,27 @@ const server = http.createServer(async (req, res) => {
       return;
     }
 
+    if (routePath === '/api/voices' && req.method === 'GET') {
+      try {
+        const { getVoices } = await getSpeech();
+        sendJSON(req, res, 200, { ok: true, voices: getVoices() });
+      } catch (err) {
+        sendJSON(req, res, 200, { ok: true, voices: [] });
+      }
+      return;
+    }
+
     if (routePath === '/api/tts' && req.method === 'POST') {
       try {
         const body = await parseBody(req);
         const text = body.text || '';
+        const voiceId = body.voiceId || null;
         if (!text) {
-                    sendJSON(req, res, 400, { error: 'No text provided' });
+          sendJSON(req, res, 400, { error: 'No text provided' });
           return;
         }
         const { synthesize } = await getSpeech();
-        const wavBuffer = await synthesize(text);
+        const wavBuffer = await synthesize(text, voiceId);
         res.writeHead(200, { 'Content-Type': 'audio/wav', 'Content-Length': wavBuffer.length });
         res.end(wavBuffer);
       } catch (err) {
@@ -558,6 +569,7 @@ const server = http.createServer(async (req, res) => {
       try {
         const body = await parseBody(req);
         const text = body.text || '';
+        const voiceId = body.voiceId || null;
         if (!text) {
           sendJSON(req, res, 400, { error: 'No text provided' });
           return;
@@ -569,7 +581,7 @@ const server = http.createServer(async (req, res) => {
           'X-Content-Type': 'audio/wav-stream',
           'Cache-Control': 'no-cache'
         });
-        for await (const wavChunk of synthesizeStream(text)) {
+        for await (const wavChunk of synthesizeStream(text, voiceId)) {
           const lenBuf = Buffer.alloc(4);
           lenBuf.writeUInt32BE(wavChunk.length, 0);
           res.write(lenBuf);
@@ -589,7 +601,7 @@ const server = http.createServer(async (req, res) => {
         const { getStatus } = await getSpeech();
         sendJSON(req, res, 200, getStatus());
       } catch (err) {
-                sendJSON(req, res, 200, { sttReady: false, ttsReady: false, sttLoading: false, ttsLoading: false });
+        sendJSON(req, res, 200, { sttReady: false, ttsReady: false, sttLoading: false, ttsLoading: false });
       }
       return;
     }

package/static/index.html

@@ -1056,6 +1056,28 @@
       border-color: var(--color-error);
     }
 
+    .voice-selector-wrapper {
+      display: flex;
+      align-items: center;
+      gap: 0.25rem;
+    }
+
+    .voice-selector {
+      padding: 0.2rem 0.5rem;
+      border: 1px solid var(--color-border);
+      border-radius: 0.375rem;
+      background: var(--color-bg-secondary);
+      color: var(--color-text-primary);
+      font-size: 0.75rem;
+      cursor: pointer;
+      max-width: 160px;
+    }
+
+    .voice-selector:focus {
+      outline: none;
+      border-color: var(--color-primary);
+    }
+
     .voice-empty {
       text-align: center;
       color: var(--color-text-secondary);
@@ -2146,6 +2168,11 @@
               <input type="checkbox" id="voiceTTSToggle" checked>
               <span>Auto-speak responses</span>
             </label>
+            <div class="voice-selector-wrapper">
+              <select class="voice-selector" id="voiceSelector" title="Select voice">
+                <option value="default">Default</option>
+              </select>
+            </div>
             <button class="voice-stop-btn" id="voiceStopSpeaking" title="Stop speaking">Stop</button>
           </div>
         </div>

package/static/js/voice.js

@@ -14,12 +14,48 @@
   var TARGET_SAMPLE_RATE = 16000;
   var spokenChunks = new Set();
   var isLoadingHistory = false;
+  var selectedVoiceId = localStorage.getItem('voice-selected-id') || 'default';
 
   function init() {
     setupTTSToggle();
     setupUI();
     setupStreamingListener();
     setupAgentSelector();
+    setupVoiceSelector();
+  }
+
+  function setupVoiceSelector() {
+    var selector = document.getElementById('voiceSelector');
+    if (!selector) return;
+    var saved = localStorage.getItem('voice-selected-id');
+    if (saved) selectedVoiceId = saved;
+    fetch(BASE + '/api/voices')
+      .then(function(res) { return res.json(); })
+      .then(function(data) {
+        if (!data.ok || !Array.isArray(data.voices)) return;
+        selector.innerHTML = '';
+        data.voices.forEach(function(voice) {
+          var opt = document.createElement('option');
+          opt.value = voice.id;
+          var label = voice.name;
+          if (voice.gender || voice.accent) {
+            var parts = [];
+            if (voice.gender) parts.push(voice.gender);
+            if (voice.accent) parts.push(voice.accent);
+            label += ' (' + parts.join(', ') + ')';
+          }
+          opt.textContent = label;
+          selector.appendChild(opt);
+        });
+        if (saved && selector.querySelector('option[value="' + saved + '"]')) {
+          selector.value = saved;
+        }
+      })
+      .catch(function() {});
+    selector.addEventListener('change', function() {
+      selectedVoiceId = selector.value;
+      localStorage.setItem('voice-selected-id', selectedVoiceId);
+    });
   }
 
   function syncVoiceSelector() {
@@ -289,7 +325,7 @@
     fetch(BASE + '/api/tts-stream', {
       method: 'POST',
       headers: { 'Content-Type': 'application/json' },
-      body: JSON.stringify({ text: text })
+      body: JSON.stringify({ text: text, voiceId: selectedVoiceId })
     }).then(function(resp) {
       if (!resp.ok) throw new Error('TTS failed');
       var reader = resp.body.getReader();