Sync voice chatId session handling

src/voice/protocol.py

@@ -18,7 +18,6 @@ from pipecat.frames.frames import (
     OutputAudioRawFrame,
     OutputTransportMessageFrame,
     OutputTransportMessageUrgentFrame,
-    TextFrame,
     TranscriptionFrame,
     UserImageRawFrame,
 )
@@ -64,13 +63,15 @@ class ProductWebsocketSerializer(FrameSerializer):
                 timestamp=frame.timestamp,
             )
 
-        if isinstance(frame, TextFrame):
-            return self._event("response.text.delta", text=frame.text)
-
+        # ProductTextStreamProcessor owns response.text.* events. TTS can also
+        # emit TextFrame subclasses internally, so serializing them here would
+        # make clients render duplicate assistant text.
         if isinstance(frame, (OutputTransportMessageFrame, OutputTransportMessageUrgentFrame)):
             if self.should_ignore_frame(frame):
                 return None
             message = frame.message
+            # Allow callers to emit a named protocol event by pushing a
+            # transport-message frame whose payload already carries a `type`.
             if isinstance(message, dict) and isinstance(message.get("type"), str):
                 event_type = message["type"]
                 payload = {k: v for k, v in message.items() if k != "type"}
@@ -99,10 +100,12 @@ class ProductWebsocketSerializer(FrameSerializer):
 
         message_type = message.get("type")
         if message_type == "session.start":
+            chat_id = message.get("chatId") or message.get("chat_id")
             return InputTransportMessageFrame(
                 message={
                     "type": "session.started",
                     "protocol": self.protocol,
+                    "chatId": chat_id if isinstance(chat_id, str) else None,
                     "audio": {
                         "encoding": "pcm_s16le",
                         "sample_rate": self._sample_rate,