Inference: Improved support for mid session TTS updates (#4412)

Author: adrian-cowham

livekit-agents/livekit/agents/inference/tts.py

@@ -502,6 +502,14 @@ async def _sentence_stream_task(ws: aiohttp.ClientWebSocketResponse) -> None:
             async for ev in sent_tokenizer_stream:
                 token_pkt = base_pkt.copy()
                 token_pkt["transcript"] = ev.token + " "
+                generation_config: dict[str, Any] = {}
+                if self._opts.voice:
+                    generation_config["voice"] = self._opts.voice
+                if self._opts.model:
+                    generation_config["model"] = self._opts.model
+                if self._opts.language:
+                    generation_config["language"] = self._opts.language
+                token_pkt["generation_config"] = generation_config
                 token_pkt["extra"] = self._opts.extra_kwargs if self._opts.extra_kwargs else {}
                 self._mark_started()
                 await ws.send_str(json.dumps(token_pkt))