Add native Qwen3 TTS streaming for Hermes

router/ai_profile_router.py

@@ -1519,6 +1519,13 @@ class Handler(BaseHTTPRequestHandler):
                 else:
                     self._send_error(503, "Sprachausgabe ist nicht installiert",
                                      "server_error", "feature_disabled")
+            elif (path == "/v1/audio/speech/pcm-stream"
+                  and self.command == "POST"):
+                if ENABLE_TTS:
+                    self._speech_pcm_stream()
+                else:
+                    self._send_error(503, "Sprachausgabe ist nicht installiert",
+                                     "server_error", "feature_disabled")
             elif path == "/v1/audio/transcriptions" and self.command == "POST":
                 if ENABLE_STT:
                     self._transcribe()
@@ -2062,6 +2069,74 @@ class Handler(BaseHTTPRequestHandler):
         self.end_headers()
         self.wfile.write(audio)
 
+    def _speech_pcm_stream(self) -> None:
+        """Pass through Qwen's native 24 kHz PCM stream without buffering."""
+        try:
+            body = self._read_body()
+            data = json.loads(body)
+        except ValueError as exc:
+            self._send_error(400, str(exc) or "ungültiges JSON",
+                             "invalid_request_error", "invalid_body")
+            return
+        if not isinstance(data, dict):
+            self._send_error(400, "Request muss ein JSON-Objekt sein",
+                             "invalid_request_error", "invalid_request")
+            return
+        text = data.get("input", data.get("text"))
+        if not isinstance(text, str) or not text.strip() or len(text) > 8000:
+            self._send_error(400, "'input' fehlt, ist leer oder zu lang",
+                             "invalid_request_error", "invalid_input")
+            return
+        voice = data.get("voice", TTS_DEFAULT_VOICE)
+        if voice not in TTS_VOICES:
+            self._send_error(400, f"ungültige Stimme: {voice!r}",
+                             "invalid_request_error", "invalid_voice")
+            return
+
+        hostport = TTS_WORKER_URL.split("://", 1)[-1]
+        host, _, port = hostport.partition(":")
+        connection = None
+        headers_sent = False
+        try:
+            connection = http.client.HTTPConnection(
+                host, int(port) if port else 80, timeout=TTS_CONNECT_TIMEOUT)
+            connection.request(
+                "POST", "/tts/pcm-stream", body=body,
+                headers={"Content-Type": "application/json",
+                         "Accept": "application/octet-stream"})
+            connection.sock.settimeout(TTS_TIMEOUT)
+            response = connection.getresponse()
+            if response.status != 200:
+                message = response.read(512).decode(errors="replace")
+                self._send_error(503,
+                                 f"TTS-Stream fehlgeschlagen: {message}",
+                                 "server_error", "tts_failed")
+                return
+            self._last_code = 200
+            self.send_response(200)
+            self.send_header("Content-Type", "application/octet-stream")
+            self.send_header("Cache-Control", "no-store")
+            self.send_header("Connection", "close")
+            self.end_headers()
+            headers_sent = True
+            while True:
+                chunk = response.read1(16384)
+                if not chunk:
+                    break
+                self.wfile.write(chunk)
+                self.wfile.flush()
+        except (OSError, http.client.HTTPException) as exc:
+            if not headers_sent:
+                try:
+                    self._send_error(502, f"TTS-Worker nicht erreichbar: {exc}",
+                                     "server_error", "tts_unavailable")
+                except (OSError, BrokenPipeError):
+                    pass
+        finally:
+            if connection is not None:
+                connection.close()
+            self.close_connection = True
+
     # ---------- Audio-Discovery ----------
 
     def _audio_models_payload(self) -> dict: