I changes the model , the same agent is able to talk over chat mode but the voice is not working?

Thanks for reaching out, Aditya. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

i was using azure LLM model

Thanks — that helps. You're seeing chat work after switching to an Azure LLM, but no voice. That usually means the agent has no active TTS configured after the model change. Quick checklist and next steps so I can give an exact fix: • Which TTS provider were you using before (if any)? • Share the code snippet you use to create the AgentSession (the lines for

llm=

and

tts=

). Wrap it in triple backticks if possible. • Any error messages in server or client logs when the agent attempts to speak? If you prefer, here’s a sample of how to explicitly set Azure TTS alongside an Azure LLM:

session = AgentSession(
    llm=openai.LLM.with_azure(...),       # your Azure LLM config
    tts=openai.TTS.with_azure(voice="ash"),  # explicit Azure TTS config
    # ... stt, vad, etc.
)

Send the snippet and any errors and I’ll provide the exact code or config fix. Sources: Realtime model plugins | LiveKit Docs | Azure OpenAI TTS plugin guide | LiveKit docs | Azure Speech TTS plugin guide | LiveKit docs

but the same LLM model is working if i connect to the agent via chat mode(text only) but as soon as i go to STT -> LLM -> TTS its not working? what could be the reason

Thanks — I understand: chat (LLM) works but the voice pipeline (STT → LLM → TTS) fails. I can help diagnose — a few targeted questions and quick checks will pinpoint the cause. Please share: • The exact

AgentSession

(or agent init) lines for

stt=

,

llm=

, and

tts=

(wrap in triple backticks). • Any error messages or warnings from server or client logs when you trigger voice. • Whether you’re using a realtime LLM or a request/response LLM (Azure realtime vs Azure standard). • Which TTS provider (Azure TTS, OpenAI TTS, etc.) and voice name you expect to use. Quick checks you can run now: • Confirm

tts=

is explicitly set in the session after switching to Azure LLM — switching LLM does not auto-enable TTS. • Look for TTS initialization or synthesis errors in logs (auth, quota, or codec errors). • If using a realtime model, verify it allows external TTS in your pipeline (some realtime setups handle audio differently). Send the code and any logs and I’ll give the exact config or fix.

TTS is working as i can her the inital greeting, looks like the response is not being sent to the LLM model, and there are no errors in the logs aswell

Thanks — I hear you: initial TTS (greeting) works, but after STT the transcript isn’t producing an LLM response (and there are no errors). I can help narrow this quickly. Please provide these small, specific items so I can pinpoint the break in the STT → LLM → TTS flow: • The exact

AgentSession

(or agent init) lines for

stt=

,

llm=

, and

tts=

(wrap in triple backticks). • Whether your Azure LLM is configured in realtime mode or standard request/response. • Do you see the STT transcript appear anywhere (client console, server logs, or agent logs)? If yes, paste one example transcript. • Any VAD or silence/utterance settings you use (e.g.,

vad

,

stop_on_silence

, timeouts). • The sequence you use to trigger the pipeline (push-to-talk vs continuous), and whether the LLM returns text when you call it directly in text mode using the same inputs. Quick checks you can run now while you gather that: • Confirm STT transcripts are reaching the process that forwards to the LLM (look for the transcript string right before the LLM call). • If possible, call the LLM directly with a sample transcript (same text STT produced) to confirm LLM responds in voice mode. • Check for any conditional logic that only forwards transcripts to LLM for chat mode (e.g., a flag that distinguishes text chat vs voice pipeline). Send the small code snippets and one transcript sample and I’ll give the exact lines to fix (or a minimal repro to run).