Overview
Keep only the main speaker and remove background voices and noise, on your servers with the VIVA SDK or in Krisp's cloud with the Krisp API.
Voice Isolation removes background voices and noise from an audio stream in real time and keeps only the main speaker. Put it first in a voice AI pipeline, before VAD, turn detection and speech-to-text, so those models hear only the user.
Choose SDK or API
| VIVA SDK | Voice Isolation API | |
|---|---|---|
| Where it runs | On your servers, on CPU, inside the VIVA SDK | In Krisp's cloud |
| How you use it | Load a model and process audio frames in your code | Stream audio over a WebSocket: wss://streaming.krisp.ai/voice-isolation |
| Languages / clients | Python, Node.js, Go, Rust, C++, C | Any WebSocket client |
| Models | Telephony, WebRTC and smart-speaker models (see Voice Isolation (VIVA SDK)) | VIVA-tel |
| Audio | Depends on the model: 8 kHz to above 16 kHz | 8,000 or 16,000 Hz |
Not sure which one fits? See SDK or API.
Key points
- The Telephony and WebRTC models remove background voices, so they expect the user to stay close to the microphone (within about 1 foot / 30 cm). For far-field audio, use the smart-speaker model, which removes noise only.
- Voice Isolation improves Turn-Taking and speech-to-text accuracy, because background speech no longer looks like the user talking.
Get started
- SDK: Quickstart: VIVA SDK, then VIVA SDK Getting Started
- API: Authentication, then the Voice Isolation API reference
Updated 9 days ago
Did this page help you?