Update app.py

app.py

@@ -109,14 +109,20 @@ def _parse_avatar_config(avatar_dir):
 def _resolve_voice(config: dict) -> str:
     """
     Pick the TTS voice from an avatar config.
-    Precedence:
-      1. edgettsname  → raw edge-tts voice ID (e.g. "en-GB-RyanNeural")
-      2. groqttsname  → legacy short name, mapped via VOICE_MAP in tts.py
-      3. fallback default
-    set_voice() in tts.py accepts either form (it detects edge-tts IDs by the "Neural" suffix
-    and falls back to the legacy mapping otherwise), so we just hand back whichever is present.
+    Precedence (first wins):
+      1. cartesiavoicename → Cartesia voice UUID (e.g. "e07c00bc-4134-...")
+      2. edgettsname       → raw edge-tts voice ID (e.g. "en-GB-RyanNeural")
+      3. groqttsname       → legacy short name, mapped via VOICE_MAP in tts.py
+      4. fallback default
+    set_voice() in tts.py detects the format and routes to the right backend:
+    UUID → Cartesia API, else → edge-tts (raw or via VOICE_MAP).
     """
-    return config.get("edgettsname") or config.get("groqttsname") or "diana"
+    return (
+        config.get("cartesiavoicename")
+        or config.get("edgettsname")
+        or config.get("groqttsname")
+        or "diana"
+    )
 
 
 def _switch_avatar(avatar_name: str):
@@ -1072,6 +1078,7 @@ const btnClear=document.getElementById('btnClear');
 // WEBSOCKET FRAME STREAMING
 // ======================================================================
 let ws=null, frameImg=new Image();
+let pendingAudio=null, pendingAudioUrl=null;
 
 function drawFrame(data){
     const blob=new Blob([data], {type:'image/jpeg'});
@@ -1083,6 +1090,15 @@ function drawFrame(data){
     frameImg.src=url;
 }
 
+function preloadAudio(url){
+    // Called from chat_stream response — gives ~2s headstart on HTTP fetch + buffer
+    // so audio is ready to play instantly when speech_start fires.
+    if(!url) return;
+    pendingAudio=new Audio(url);
+    pendingAudioUrl=url;
+    pendingAudio.load();
+}
+
 function connectWebSocket(){
     const proto=location.protocol==='https:'?'wss:':'ws:';
     ws=new WebSocket(proto+'//'+location.host+'/ws/video');
@@ -1094,15 +1110,20 @@ function connectWebSocket(){
         if(typeof e.data==='string'){
             const msg=JSON.parse(e.data);
             if(msg.type==='speech_start' && msg.audio_url){
-                // Preload audio then play — frames draw continuously
-                const audio=new Audio(msg.audio_url);
-                audio.addEventListener('canplaythrough', ()=>{
-                    audio.play().catch(err=>console.warn('[AUDIO]',err));
-                }, {once:true});
-                audio.load();
+                if(pendingAudio && pendingAudioUrl===msg.audio_url){
+                    // Audio was preloaded after /api/chat/stream returned — play instantly
+                    pendingAudio.play().catch(err=>console.warn('[AUDIO]',err));
+                    pendingAudio=null; pendingAudioUrl=null;
+                } else {
+                    // Fallback (greeting / no preload) — load + wait for canplaythrough
+                    const audio=new Audio(msg.audio_url);
+                    audio.addEventListener('canplaythrough', ()=>{
+                        audio.play().catch(err=>console.warn('[AUDIO]',err));
+                    }, {once:true});
+                    audio.load();
+                }
             }
         } else {
-            // Always draw frames immediately — server paces at 25fps
             drawFrame(e.data);
         }
     };
@@ -1158,6 +1179,10 @@ async function sendMessage(){
         typing.contentEl.classList.remove('typing-dots');
         typing.contentEl.textContent=data.text;
 
+        // Start fetching the TTS audio NOW — gives ~2s headstart so it's
+        // buffered and ready to play instantly when WS emits speech_start.
+        if(data.audio_url) preloadAudio(data.audio_url);
+
         // Audio + lipsync frames arrive via WebSocket automatically
         // Wait for speech duration before allowing next message
         if(data.audio_duration){