How can we use Deepgram endpointing instead of Silero VAD?

pipecat-ai / pipecat

Open Source framework for voice and multimodal conversational AI

BSD 2-Clause "Simplified" License

3.3k stars 309 forks source link

How can we use Deepgram endpointing instead of Silero VAD? #525

Open nisalr opened 4 weeks ago

nisalr commented 4 weeks ago

Sometimes Silero detects that the user is speaking when Deepgram doesn't (and vice versa). Is there a way to just use Deepgram endpointing so that these inconsistencies don't occur?

manish-baghel commented 3 weeks ago

looks like we need to push UserStartedSpeaking and UserStoppedSpeakingFrame from stt layer and disable the vad in transport layer