when using voiceagent and have `def prewarm(proc: ...
# ask-ai
s
when using voiceagent and have
def prewarm(proc: JobProcess):
proc.userdata["vad"] = silero.VAD.load(
min_speech_duration=0.25,
min_silence_duration=1.0,
prefix_padding_duration=0.3,
activation_threshold=0.6,
sample_rate=16000,
force_cpu=True,
)
and
agent = Assistant()
session = AgentSession(
vad=ctx.proc.userdata["vad"],
# vad=silero.VAD.load(),
stt=sst_openai,
llm=llm_openai,
tts=tts_openai,
turn_detection=MultilingualModel(),
)
, lantency is very long # Problem: seeing too many VAD inferences / too frequent VAD triggering in LiveKit’s voice pipeline. # If Silero VAD is being triggered too frequently, you need to throttle the VAD input or adjust Silero parameters.
t
Thanks for reaching out, CellCS. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
I'm sorry, but I don't have the information you need at the moment. If you would like a member of the LiveKit team to respond, please ask your question in one of the other channels.