[TTS] Add missing _generate_pcm_chunks for OmniOpenAIServingSpeech streaming (#2569)

Signed-off-by: Yueqian Lin <pandaleefree@gmail.com>
Signed-off-by: Yueqian Lin <70319226+linyueqian@users.noreply.github.com>
Co-authored-by: vveerrgg <vveerrgg@users.noreply.github.com>

tests/entrypoints/openai_api/test_serving_speech_stream.py

@@ -385,3 +385,18 @@ class TestStreamingSpeechWebSocket:
 
         speech_service.engine_client.abort.assert_awaited_once_with("req-abort")
         assert websocket.send_json.await_count == 2
+
+
+class TestGeneratePcmChunksContract:
+    """Guard: _generate_pcm_chunks must exist on OmniOpenAIServingSpeech.
+
+    The WebSocket handler calls speech_service._generate_pcm_chunks()
+    at runtime. If the method is removed, all WS TTS streaming breaks
+    with an AttributeError. This test catches that at CI time.
+    """
+
+    def test_generate_pcm_chunks_defined(self):
+        assert hasattr(OmniOpenAIServingSpeech, "_generate_pcm_chunks")
+        assert asyncio.iscoroutinefunction(OmniOpenAIServingSpeech._generate_pcm_chunks) or callable(
+            OmniOpenAIServingSpeech._generate_pcm_chunks
+        )

vllm_omni/entrypoints/openai/serving_speech.py

@@ -1459,6 +1459,15 @@ class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
         )
         return request_id, generator, tts_params
 
+    async def _generate_pcm_chunks(self, generator, request_id: str):
+        """Yield raw PCM byte chunks from the engine generator.
+
+        Delegates to ``_generate_audio_chunks`` with ``response_format="pcm"``.
+        Used by the WebSocket streaming handler and ``_iter_pcm_audio_bytes``.
+        """
+        async for chunk in self._generate_audio_chunks(generator, request_id, response_format="pcm"):
+            yield chunk
+
     async def _iter_pcm_audio_bytes(self, request: OpenAICreateSpeechRequest):
         """Yield raw PCM bytes for a speech request as soon as chunks are decoded."""
         request_id, generator, _ = await self._prepare_speech_generation(request)