Audio Settings
Choose the caller's input audio filter with audio.noiseCancellation in an Advanced Settings resource.
Overview
The filter processes incoming audio before voice activity detection and transcription. It applies to both LLM + TTS + STT pipelines and STS (Speech-to-Speech) models. Use it to reduce background sounds and nearby voices that could trigger false turns or affect recognition.
Data model
| Field | Type | Default | Description |
|---|---|---|---|
noiseCancellation | string | null | Based on call type | Filter applied to the caller's audio before processing |
noiseCancellation accepts the following values. Any other value is rejected with a 400 response. Omitting the field or setting it to null picks the filter from the call type: bvc_telephony for phone calls over SIP, bvc for WebRTC calls.
| Value | Removes | Best for |
|---|---|---|
bvc_telephony | Background noise and nearby voices (TV, colleagues), tuned for narrowband phone audio | Phone calls over SIP (default for SIP) |
bvc | Background noise and nearby voices (TV, colleagues), isolating the primary speaker | WebRTC calls with wideband audio (default for WebRTC) |
nc | Non-voice noise only, every voice in the room stays audible | Callers who share a microphone with other speakers |
disabled | Nothing | Studio-quality input, or debugging audio issues |
An explicit value selects the same filter for every call type. Set "nc" to remove non-voice noise while retaining other speakers. The request path is relative to the Data API server URL in the API reference, which includes /data_next:
Code
Code
Note
Tune VAD Settings alongside noise cancellation. If the Assistant still reacts to background sounds after filtering, raise activationThreshold or minSpeechDuration. Setting noiseCancellation to "disabled" usually calls for a higher activationThreshold to compensate.