gemma

core/filler_audio.py

@@ -107,10 +107,28 @@ class FillerAudioLibrary:
         segments: Iterable[tuple[str, str | None]],
     ) -> bytes:
         clip = bytearray()
-        for text, language_code in segments:
+        for i, (text, language_code) in enumerate(segments):
+            LOGGER.info(
+                "synthesize_segments: starting segment %s language=%s text=%r",
+                i,
+                language_code,
+                text[:60],
+            )
             segment = await self._synthesize_clip(tts, text, language_code=language_code)
             if segment:
                 clip.extend(segment)
+                LOGGER.info(
+                    "synthesize_segments: completed segment %s bytes=%s total_bytes=%s",
+                    i,
+                    len(segment),
+                    len(clip),
+                )
+            else:
+                LOGGER.warning(
+                    "synthesize_segments: segment %s produced NO audio; language=%s",
+                    i,
+                    language_code,
+                )
         return bytes(clip)
 
     def _load_from_files(self) -> None:

core/session.py

@@ -1868,6 +1868,7 @@ class CallSession:
                     len(segment_text),
                     _preview_text(segment_text),
                 )
+                audio_segment_chunk_count = 0
                 async for audio_chunk in self._tts.synthesize_stream(
                     one_shot_text_stream(),
                     language_code=language_code,
@@ -1875,6 +1876,7 @@ class CallSession:
                     self._ensure_generation(epoch)
                     if not audio_chunk:
                         continue
+                    audio_segment_chunk_count += 1
                     audio_chunk_count += 1
                     audio_byte_count += len(audio_chunk)
                     if not first_audio_seen:
@@ -1887,6 +1889,20 @@ class CallSession:
                         )
                         self._set_state(SessionState.ASSISTANT_SPEAKING, reason="initial greeting started")
                     await self.transport.send_audio(audio_chunk)
+                
+                if audio_segment_chunk_count == 0:
+                    LOGGER.warning(
+                        "realtime session %s initial greeting segment produced NO audio: language=%s",
+                        self.session_id,
+                        language_code,
+                    )
+                else:
+                    LOGGER.info(
+                        "realtime session %s initial greeting segment completed: language=%s chunks=%s",
+                        self.session_id,
+                        language_code,
+                        audio_segment_chunk_count,
+                    )
             await self.transport.flush_audio()
             self._ensure_generation(epoch)
             LOGGER.info(

main.py

@@ -103,24 +103,20 @@ def _initial_greeting_segments() -> tuple[tuple[str, str | None], ...]:
         language = str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_LANGUAGE_CODE", "")).strip() or None
         return ((raw, language),)
 
-    ru_text = str(
+    ru_text = (
-        os.getenv(
+        str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_RU_TEXT") or "").strip()
-            "REALTIME_VOICE_INITIAL_GREETING_RU_TEXT",
+        or (
-            (
             "Здравствуйте! Меня зовут Айнур, я эй-ай-ассистент компании ДиджиОпс. "
             "Как я могу к вам обращаться и на каком языке вам удобнее продолжить: на русском или на казахском?"
-            ),
         )
-    ).strip()
+    )
-    kk_text = str(
+    kk_text = (
-        os.getenv(
+        str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_KK_TEXT") or "").strip()
-            "REALTIME_VOICE_INITIAL_GREETING_KK_TEXT",
+        or (
-            (
             "Сәлеметсіз бе! Менің атым Айнұр, мен ДиджиОпс компаниясының эй-ай ассистентімін. "
             "Сізге қалай жүгінсем болады және қай тілде сөйлескен ыңғайлы: орысша ма, қазақша ма?"
-            ),
         )
-    ).strip()
+    )
     segments: list[tuple[str, str | None]] = []
     if ru_text:
         segments.append((ru_text, "ru"))

providers/tts.py

@@ -192,7 +192,11 @@ class ElevenLabsTTS(BaseTTS):
         if not self._voice_id:
             raise RuntimeError("ELEVENLABS_TTS_VOICE_ID is required for ElevenLabs TTS")
 
-        effective_language_code = (str(language_code).strip() if language_code else "") or self._language_code
+        # If language_code is explicitly provided, use it. 
+        # If it's None, use the instance default self._language_code.
+        # To trigger auto-detection, language_code should be an empty string or the default should be None.
+        effective_language_code = language_code if language_code is not None else self._language_code
+        
         if self._resolved_transport_mode() == "http_stream":
             async for audio_chunk in self._synthesize_http_stream(
                 text_stream,
@@ -234,7 +238,7 @@ class ElevenLabsTTS(BaseTTS):
             self._output_format,
             self._provider_sample_rate_hz,
             self._target_sample_rate_hz,
-            language_code,
+            language_code or "auto",
             self._auto_mode,
         )
         try:
@@ -297,13 +301,14 @@ class ElevenLabsTTS(BaseTTS):
                             if audio_chunk:
                                 yield audio_chunk
                         if self._is_error_payload(payload):
+                            error_detail = self._format_error_payload(payload)
                             LOGGER.error(
-                                "ElevenLabs TTS error payload: voice_id=%s model=%s payload=%s",
+                                "ElevenLabs TTS error payload: voice_id=%s model=%s error=%s",
                                 self._voice_id,
                                 self._model_id,
-                                self._format_error_payload(payload),
+                                error_detail,
                             )
-                            raise RuntimeError(self._format_error_payload(payload))
+                            raise RuntimeError(error_detail)
                         if bool(payload.get("isFinal")) or bool(payload.get("is_final")):
                             LOGGER.info(
                                 "ElevenLabs TTS final payload: voice_id=%s model=%s chunks=%s "
@@ -327,9 +332,9 @@ class ElevenLabsTTS(BaseTTS):
         except asyncio.TimeoutError as exc:
             raise RuntimeError("ElevenLabs TTS WebSocket request timed out") from exc
         except WebSocketException as exc:
-            raise RuntimeError("ElevenLabs TTS WebSocket stream failed") from exc
+            raise RuntimeError(f"ElevenLabs TTS WebSocket stream failed: {exc}") from exc
         except OSError as exc:
-            raise RuntimeError("ElevenLabs TTS WebSocket connection failed") from exc
+            raise RuntimeError(f"ElevenLabs TTS WebSocket connection failed: {exc}") from exc
 
     async def _synthesize_http_stream(
         self,
@@ -364,7 +369,7 @@ class ElevenLabsTTS(BaseTTS):
             self._output_format,
             self._provider_sample_rate_hz,
             self._target_sample_rate_hz,
-            language_code,
+            language_code or "auto",
         )
 
         try:
@@ -389,9 +394,9 @@ class ElevenLabsTTS(BaseTTS):
                     async with session.post(url, headers=headers, json=payload) as response:
                         if response.status >= 400:
                             payload_text = await response.text()
-                            raise RuntimeError(
+                            error_msg = f"ElevenLabs TTS HTTP {response.status}: {payload_text[:300]}"
-                                f"ElevenLabs TTS HTTP {response.status}: {payload_text[:300]}"
+                            LOGGER.error(error_msg)
-                            )
+                            raise RuntimeError(error_msg)
                         async for audio_chunk in response.content.iter_chunked(4096):
                             if not audio_chunk:
                                 continue
@@ -439,7 +444,7 @@ class ElevenLabsTTS(BaseTTS):
         except asyncio.TimeoutError as exc:
             raise RuntimeError("ElevenLabs TTS HTTP stream timed out") from exc
         except aiohttp.ClientError as exc:
-            raise RuntimeError("ElevenLabs TTS HTTP stream failed") from exc
+            raise RuntimeError(f"ElevenLabs TTS HTTP stream failed: {exc}") from exc
 
         LOGGER.info(
             "ElevenLabs TTS HTTP stream completed: voice_id=%s model=%s requests=%s chunks=%s "
@@ -462,9 +467,9 @@ class ElevenLabsTTS(BaseTTS):
             "sync_alignment": "false",
             "apply_text_normalization": self._apply_text_normalization,
         }
-        effective_language = (str(language_code).strip() if language_code else "") or self._language_code
+        # Use provided language_code or fallback to None for auto-detection
-        if effective_language:
+        if language_code:
-            query["language_code"] = effective_language
+            query["language_code"] = language_code
         encoded_voice_id = quote(self._voice_id, safe="")
         return f"{self._ws_base}/v1/text-to-speech/{encoded_voice_id}/stream-input?{urlencode(query)}"
 
@@ -482,9 +487,9 @@ class ElevenLabsTTS(BaseTTS):
             "voice_settings": self._voice_settings,
             "apply_text_normalization": self._apply_text_normalization,
         }
-        effective_language = (str(language_code).strip() if language_code else "") or self._language_code
+        # Use provided language_code or fallback to None for auto-detection
-        if effective_language:
+        if language_code:
-            payload["language_code"] = effective_language
+            payload["language_code"] = language_code
         return payload
 
     def _resolved_transport_mode(self) -> str: