#Agent is picking up noise and responding
old quest
For reference. I am using Gemini 2.0 flash and voice settings are:
TTS output format-16000
Optimize streaming latency-3
stability-0.50
Speed 1.05
Stability0.80
vivid kiln
same issue here, agent is picking up background noise / conversation and transcribing it
stable pagoda
The issue isn't related to the LLM model or the settings; it's all about the audio you're providing. If the audio is too noisy, the turn-taking won't work reliably. The AI will keep interrupting or hear noise that it interprets as something it should respond to, which isn't ideal.
Would it be possible to run noise cancellation on your end before sending the information to ElevenLabs? We're also working on releasing a new turn-taking model soon that should perform better. Unfortunately, I don't have much information on this and can't look it up right now, but we are working on a V2 conversational AI model.
old quest
OK. just so you know, we are using custom voice. We are not using elevenlabs voices, could this be the problem?
vivid kiln
We are using convAI with twilio so impossible tu run something in between…
lofty hinge
OK. just so you know, we are using custom voice. We are not using elevenlabs voi...
Using a Professional Voice Clone will not have any impact on this. As Jonathan said, you'll need to run some background noise cancellation on your end, we are working on finding a solution to this but everyone's "background noise" is different, different sounds (road noise for Trucker dispatch services or music/other people talking in the background for example) so it's hard to find a one stop solution for all use cases.
old quest
@lofty hinge . When we listen to the call recordings, we don't get to hear anything but can see .... in transcript.
I cannot listen to any noise in the audio
but its there in the transcript
This is happening with multiple calls. It interrupts tool call and is very frustrating for the callers