I'm using LiveKit Agents Python framework with Age...
# ask-aic
chilly-nest-87883
07/27/2025, 9:44 PMI'm using LiveKit Agents Python framework with AgentSession and I want to access interim transcripts from Deepgram STT for custom interruption
logic. You mentioned that Deepgram emits INTERIM_TRANSCRIPT and FINAL_TRANSCRIPT events, but how do I actually listen to these events when using the
high-level AgentSession API?
My current code structure is:
session = AgentSession(
stt=deepgram.STT(model="nova-3", interim_results=True),
llm=openai.LLM(),
tts=cartesia.TTS(),
vad=silero.VAD.load(),
turn_detection=EnglishModel()
)
await session.start(agent=agent, room=room)
I need to intercept interim transcripts as they arrive (before final transcription) to implement word-based interruption logic. What's the correct
way to access the STT stream events or interim transcript data within the AgentSession framework? Should I be using session event handlers, or do I
need to access the STT instance directly?
t
tall-airline-16756
07/27/2025, 9:45 PMThanks for reaching out, Clovis. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
tall-airline-16756
07/27/2025, 9:46 PMTo access interim transcripts from Deepgram STT while using the high-level
AgentSessioninterim_results=True Copy code
session = AgentSession(
stt=deepgram.STT(model="nova-3", interim_results=True),
llm=openai.LLM(),
tts=cartesia.TTS(),
vad=silero.VAD.load(),
turn_detection=EnglishModel()
)on_interim_transcripton_final_transcript Copy code
async def on_interim_transcript(self, ev: stt.SpeechEvent):
interim_text = ev.alternatives[0].text
print(f"Interim: {interim_text}")
# Add your custom interruption logic here
async def on_final_transcript(self, ev: stt.SpeechEvent):
final_text = ev.alternatives[0].text
print(f"Final: {final_text}")AgentSessioncurrent_transcript Copy code
class MyAgent:
def __init__(self):
self.session = AgentSession(
stt=deepgram.STT(model="nova-3", interim_results=True),
llm=openai.LLM(),
tts=cartesia.TTS(),
vad=silero.VAD.load(),
turn_detection=EnglishModel()
)
self.session.on_interim_transcript = self.on_interim_transcript
self.session.on_final_transcript = self.on_final_transcript
async def on_interim_transcript(self, ev: stt.SpeechEvent):
interim_text = ev.alternatives[0].text
print(f"Interim: {interim_text}")
async def on_final_transcript(self, ev: stt.SpeechEvent):
final_text = ev.alternatives[0].text
print(f"Final: {final_text}")
async def start(self, agent, room):
await self.session.start(agent=agent, room=room)c
chilly-nest-87883
07/27/2025, 10:43 PMI have registered interim transcript handlers using session.on_interim_transcript = my_handler but the handler never fires when users speak. The
logs show that Deepgram STT is configured with interim_results=True and smart_format=True, and I can see final transcript events in
conversation_item_added, but my interim/final transcript handlers never receive events. What's the correct way to register and receive interim
transcript events in LiveKit Agents? Do I need to use a different event registration method or additional configuration?
t
tall-airline-16756
07/27/2025, 10:44 PMHey there! 👋 Dealing with some tricky
session.on_interim_transcriptuser_transcriptioninterim_results=True5 Views