Fix bilingual initial greeting playback

core/session.py

@@ -1840,30 +1840,50 @@ class CallSession:
     ) -> None:
         first_audio_seen = False
         started_monotonic = time.perf_counter()
-        cached_greeting = (
-            self._filler_audio.pick("initial_greeting", allow_fallback=False)
-            if self._filler_audio is not None
-            else None
-        )
-        if cached_greeting:
-            await self._play_cached_initial_greeting(
-                epoch=epoch,
-                greeting_text=greeting_text,
-                pcm_audio=cached_greeting,
-                started_monotonic=started_monotonic,
-            )
-            return
         try:
             audio_chunk_count = 0
             audio_byte_count = 0
-            for segment_text, language_code in greeting_segments:
+            for segment_index, (segment_text, language_code) in enumerate(greeting_segments, start=1):
+                cached_segment = (
+                    self._filler_audio.pick(
+                        f"initial_greeting_{segment_index}_{language_code or 'default'}",
+                        allow_fallback=False,
+                    )
+                    if self._filler_audio is not None
+                    else None
+                )
+                if cached_segment:
+                    self._ensure_generation(epoch)
+                    audio_chunk_count += 1
+                    audio_byte_count += len(cached_segment)
+                    if not first_audio_seen:
+                        first_audio_seen = True
+                        self._log_latency(
+                            "initial_greeting_cached_ttfa",
+                            started_monotonic,
+                            epoch=epoch,
+                            message="session start -> cached initial greeting first audio",
+                        )
+                        self._set_state(SessionState.ASSISTANT_SPEAKING, reason="cached initial greeting started")
+                    LOGGER.info(
+                        "realtime session %s initial greeting cached segment: epoch=%s index=%s language=%s bytes=%s",
+                        self.session_id,
+                        epoch,
+                        segment_index,
+                        language_code,
+                        len(cached_segment),
+                    )
+                    await self.transport.send_audio(cached_segment)
+                    continue
+
                 async def one_shot_text_stream() -> AsyncGenerator[str, None]:
                     yield segment_text
 
                 LOGGER.info(
-                    "realtime session %s initial greeting segment start: epoch=%s language=%s chars=%s text=%r",
+                    "realtime session %s initial greeting segment start: epoch=%s index=%s language=%s chars=%s text=%r",
                     self.session_id,
                     epoch,
+                    segment_index,
                     language_code,
                     len(segment_text),
                     _preview_text(segment_text),

main.py

@@ -97,25 +97,32 @@ def _single_initial_greeting_enabled() -> bool:
     return raw in {"1", "true", "yes", "on"}
 
 
+def _env_text_or_default(name: str, default: str) -> str:
+    value = os.getenv(name)
+    if value is None or not str(value).strip():
+        return default
+    return str(value).strip()
+
+
 def _initial_greeting_segments() -> tuple[tuple[str, str | None], ...]:
     raw = str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_TEXT", "")).strip()
     if raw and _single_initial_greeting_enabled():
         language = str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_LANGUAGE_CODE", "")).strip() or None
         return ((raw, language),)
 
-    ru_text = (
-        str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_RU_TEXT") or "").strip()
-        or (
+    ru_text = _env_text_or_default(
+        "REALTIME_VOICE_INITIAL_GREETING_RU_TEXT",
+        (
             "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?"
+        ),
     )
-    )
-    kk_text = (
-        str(os.getenv("REALTIME_VOICE_INITIAL_GREETING_KK_TEXT") or "").strip()
-        or (
+    kk_text = _env_text_or_default(
+        "REALTIME_VOICE_INITIAL_GREETING_KK_TEXT",
+        (
             "Сәлеметсіз бе! Менің атым Айнұр, мен ДиджиОпс компаниясының эй-ай ассистентімін. "
             "Сізге қалай жүгінсем болады және қай тілде сөйлескен ыңғайлы: орысша ма, қазақша ма?"
-        )
+        ),
     )
     segments: list[tuple[str, str | None]] = []
     if ru_text:
@@ -213,12 +220,32 @@ class RealtimeVoiceService:
                 len(greeting_text),
                 greeting_text[:120],
             )
-            clip = await self._filler_audio.synthesize_segments(self._tts, greeting_segments)
-            if not clip:
-                LOGGER.warning("initial greeting audio preload produced empty clip")
+            total_bytes = 0
+            for index, (segment_text, language_code) in enumerate(greeting_segments, start=1):
+                segment_clip = await self._filler_audio.synthesize_segments(
+                    self._tts,
+                    ((segment_text, language_code),),
+                )
+                if not segment_clip:
+                    LOGGER.warning(
+                        "initial greeting segment preload produced empty clip: index=%s language=%s",
+                        index,
+                        language_code,
+                    )
+                    continue
+                clip_key = f"initial_greeting_{index}_{language_code or 'default'}"
+                self._filler_audio.add_clip(clip_key, segment_clip)
+                total_bytes += len(segment_clip)
+                LOGGER.info(
+                    "initial greeting segment preload done: key=%s language=%s bytes=%s",
+                    clip_key,
+                    language_code,
+                    len(segment_clip),
+                )
+            if total_bytes <= 0:
+                LOGGER.warning("initial greeting audio preload produced no segment clips")
                 return
-            self._filler_audio.add_clip("initial_greeting", clip)
-            LOGGER.info("initial greeting audio preload done: bytes=%s", len(clip))
+            LOGGER.info("initial greeting audio preload done: bytes=%s", total_bytes)
         except Exception:
             LOGGER.exception("failed to preload initial greeting audio")