Thanks — I hear you. You’re asking whether others see voice AI failing or degrading in noisy environments or when there are background speakers, and the answer is: yes, this is a common problem. Brief summary of what's happening • Background noise (traffic, typing, HVAC) and other people talking make speech-to-text and voice agents less accurate. • Multiple speakers or background conversations are especially problematic because standard noise cancellation often treats other voices as signal rather than noise. • Aggressive noise-cancellation models can sometimes remove valid speech in poor audio conditions, and very short utterances (1–2 words) are harder to detect reliably. What you can try right away • Use Background Voice Cancellation (BVC) instead of standard noise-cancellation for agent-side processing — BVC is designed to remove other speakers and noise while keeping the primary speaker’s voice clear. • For telephony/SIP calls use

BVCTelephony()

which is tuned for narrow-band audio. • Avoid double-processing the audio: if you apply agent-side BVC, disable client-side Krisp/SIP NC to prevent artifacts that hurt recognition. • Temporarily disable noise cancellation to check whether NC is suppressing speech; that helps confirm whether NC is causing the degradation. • Consider dynamic model selection: apply different filtering models for WebRTC participants vs SIP participants to match the audio bandwidth and channel conditions. • Be cautious with very short utterances — you may need to tune detection thresholds or allow more context for speech models. If you want, I can help troubleshoot your specific setup • Tell me whether your participants are WebRTC or SIP, and whether you’re using agent-side BVC or any client-side noise cancellation (Krisp, hardware NC, etc.). • Share an example: short/long utterances failing, or specific noisy scenarios (office chatter, music, phone hold music). • If you have logs or timestamps of failed transcriptions, those help narrow root causes. Would you like to walk through a targeted checklist for your environment (WebRTC vs SIP) or try a quick test configuration I can outline? Sources: Choosing the right noise cancellation for your agent | LiveKit | Troubleshooting STT not picking up utterances | LiveKit