feat: Phase 2 — Audio STT/TTS

- TTS: edge-tts 7.2.8 (fr-FR-DeniseNeural) — génération MP3 asynchrone
- STT: Web Speech API navigateur (SpeechRecognition fr-FR)
- Audio playback: base64 MP3 décodé et joué via Audio API
- Bouton micro avec animation pulse pendant l'écoute
- Chemin absolu edge-tts (systemd PATH)
- Audio envoyé dans le message 'response' existant (champ audio)

Coût: latence TTS ~7s, zéro dépendance serveur pour le STT.

bot_direct.py

@@ -6,10 +6,12 @@ PoC minimal : WebSocket texte → API Hermes → réponse texte.
 Pas de Pipecat — intégration directe pour valider le concept.
 """
 import asyncio
+import base64
 import json
 import os
 import subprocess
 import sys
+import tempfile
 import httpx
 from websockets.asyncio.server import serve
 from websockets.http11 import Response, Headers
@@ -42,6 +44,33 @@ SYSTEM_PROMPT = (
     "Pas de markdown. Réponds comme si tu parlais à l'oral."
 )
 
+# ─── TTS (Text-to-Speech) ─────────────────────────────────────
+EDGE_TTS_VOICE = "fr-FR-DeniseNeural"  # Voix féminine française naturelle
+EDGE_TTS_BIN = "/home/openclaw/workspace/jarvis-pipecat/venv/bin/edge-tts"
+
+async def generate_tts(text: str) -> bytes:
+    """Génère un MP3 à partir du texte avec edge-tts. Retourne les bytes audio."""
+    with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as f:
+        mp3_path = f.name
+    try:
+        proc = await asyncio.create_subprocess_exec(
+            EDGE_TTS_BIN, "--voice", EDGE_TTS_VOICE,
+            "--text", text, "--write-media", mp3_path,
+            stdout=asyncio.subprocess.DEVNULL,
+            stderr=asyncio.subprocess.DEVNULL,
+        )
+        await asyncio.wait_for(proc.wait(), timeout=15)
+        with open(mp3_path, "rb") as f:
+            return f.read()
+    except Exception as e:
+        print(f"⚠️  TTS error: {e}")
+        return b""
+    finally:
+        try:
+            os.unlink(mp3_path)
+        except:
+            pass
+
 # ─── Page HTML ───────────────────────────────────────────────
 HTML_PAGE = """<!DOCTYPE html>
 <html lang="fr">
@@ -71,6 +100,9 @@ button{padding:10px 20px;border-radius:8px;border:none;background:var(--cyan);
   color:var(--bg);font-weight:bold;cursor:pointer;font-size:.95em}
 button:hover{opacity:.9}
 button:disabled{opacity:.5;cursor:default}
+#mic{width:40px;height:40px;font-size:1.2em;flex-shrink:0;padding:0;display:flex;align-items:center;justify-content:center;border-radius:50%}
+#mic.listening{background:var(--red);animation:pulse 1s infinite}
+@keyframes pulse{0%,100%{box-shadow:0 0 0 0 var(--red)}50%{box-shadow:0 0 0 8px transparent}}
 </style>
 </head>
 <body>
@@ -78,14 +110,17 @@ button:disabled{opacity:.5;cursor:default}
 <div id="status" class="off">🔴 Déconnecté</div>
 <div id="chat"></div>
 <div class="row">
-  <input id="inp" placeholder="Pose ta question..." autofocus>
+  <input id="inp" placeholder="Pose ta question ou parle...">
+  <button id="mic" title="Parler">🎤</button>
   <button id="btn">Envoyer</button>
 </div>
 <script>
 const s=document.getElementById('status'),c=document.getElementById('chat'),
-  i=document.getElementById('inp'),b=document.getElementById('btn');
-let ws=null,recon=0,reconMax=20,reconDelay=2000;
-function m(t,r){const d=document.createElement('div');d.className='msg '+r;d.textContent=t;c.appendChild(d);c.scrollTop=c.scrollHeight}
+  i=document.getElementById('inp'),b=document.getElementById('btn'),
+  mic=document.getElementById('mic');
+let ws=null,recon=0,reconMax=20,reconDelay=2000,audioCtx=null;
+
+function addMsg(t,r){const d=document.createElement('div');d.className='msg '+r;d.textContent=t;c.appendChild(d);c.scrollTop=c.scrollHeight}
 function connect(){ws=new WebSocket((location.protocol==='https:'?'wss:':'ws:')+'//'+location.host+'/ws');
 ws.onopen=()=>{recon=0;reconDelay=2000;s.textContent='🟢 Connecté — JARVIS v3 prêt';s.className='on';b.disabled=false;i.focus()};
 ws.onclose=()=>{b.disabled=true;if(recon<reconMax){recon++;let d=Math.min(reconDelay*Math.pow(1.5,recon-1),30000);
@@ -93,12 +128,34 @@ s.textContent='🟡 Reconnexion '+recon+'/'+reconMax+' dans '+Math.round(d/1000)
 else{s.textContent='🔴 Connexion perdue — recharge la page';s.className='off'}};
 ws.onerror=()=>ws.close();
 ws.onmessage=e=>{try{let d=JSON.parse(e.data);if(ws._tid){clearTimeout(ws._tid);ws._tid=null}
-if(d.type==='welcome')m(d.content,'sys');
-else if(d.type==='response')m(d.content,'jarvis');
+if(d.type==='welcome')addMsg(d.content,'sys');
+else if(d.type==='response'){addMsg(d.content,'jarvis');if(d.audio)playAudio(d.audio)}
 else if(d.type==='thinking')s.textContent='🤔 JARVIS réfléchit...';
-else if(d.type==='error')m('❌ '+d.content,'sys')}catch(x){m(e.data,'jarvis')};
-s.textContent='🟢 Connecté — JARVIS v3 prêt';b.disabled=false}}
-function send(){let t=i.value.trim();if(!t||!ws||ws.readyState!==WebSocket.OPEN)return;m(t,'user');ws.send(JSON.stringify({text:t}));i.value='';b.disabled=true;s.textContent='🤔 JARVIS réfléchit...';let tid=setTimeout(()=>{if(b.disabled){b.disabled=false;s.textContent='⚠️ JARVIS prend du temps... tu peux réessayer'}},25000);ws._tid=tid}
+else if(d.type==='error')addMsg('❌ '+d.content,'sys')}catch(x){addMsg(e.data,'jarvis')};
+s.textContent='🟢 Connecté — JARVIS v3 prêt';b.disabled=false;mic.disabled=false}}
+
+function send(){let t=i.value.trim();if(!t||!ws||ws.readyState!==WebSocket.OPEN)return;addMsg(t,'user');ws.send(JSON.stringify({text:t}));i.value='';b.disabled=true;mic.disabled=true;s.textContent='🤔 JARVIS réfléchit...';let tid=setTimeout(()=>{if(b.disabled){b.disabled=false;mic.disabled=false;s.textContent='⚠️ JARVIS prend du temps... tu peux réessayer'}},25000);ws._tid=tid}
+
+// ── Audio playback ──
+function playAudio(b64){try{const bin=atob(b64);const bytes=new Uint8Array(bin.length);for(let i=0;i<bin.length;i++)bytes[i]=bin.charCodeAt(i);const blob=new Blob([bytes],{type:'audio/mp3'});const url=URL.createObjectURL(blob);const a=new Audio(url);a.onended=()=>URL.revokeObjectURL(url);a.play().catch(e=>console.log('Audio play:',e))}catch(e){console.log('Audio error:',e)}}
+
+// ── Speech Recognition ──
+const SpeechRecognition=window.SpeechRecognition||window.webkitSpeechRecognition;
+let rec=null,listening=false;
+if(SpeechRecognition){
+  rec=new SpeechRecognition();rec.lang='fr-FR';rec.interimResults=false;rec.continuous=false;
+  rec.onresult=e=>{const t=e.results[0][0].transcript.trim();if(t){i.value=t;send()}};
+  rec.onerror=e=>{console.log('STT error:',e.error);stopListening()};
+  rec.onend=()=>stopListening();
+}
+function toggleMic(){
+  if(!rec){alert('Reconnaissance vocale non supportée sur ce navigateur');return}
+  if(listening){rec.stop();return}
+  listening=true;mic.classList.add('listening');mic.textContent='🔴';s.textContent='🎤 Je t\'écoute...';
+  try{rec.start()}catch(e){stopListening()}
+}
+function stopListening(){listening=false;mic.classList.remove('listening');mic.textContent='🎤';if(s.textContent.includes('écoute'))s.textContent='🟢 Connecté — JARVIS v3 prêt'}
+mic.onclick=toggleMic;
 b.onclick=send;i.onkeydown=e=>{if(e.key==='Enter')send()};connect()
 </script>
 </body>
@@ -177,9 +234,14 @@ async def ws_handler(websocket):
                 messages.append({"role": "assistant", "content": reply})
                 print(f"🤖 {reply[:80]}")
                 
+                # Générer TTS en parallèle de l'envoi texte
+                audio_bytes = await generate_tts(reply)
+                audio_b64 = base64.b64encode(audio_bytes).decode() if audio_bytes else ""
+                
                 await websocket.send(json.dumps({
                     "type": "response",
                     "content": reply,
+                    "audio": audio_b64,
                 }))
                 
             except Exception as e: