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(