Background noise breaks speech recognition, turn detection and interruptions. Clean the audio on the user's device before it reaches your agent.
A TV, a café or a second speaker makes the agent hear words nobody said, interrupt itself or wait for a pause that never comes. Removing noise at the source fixes the input for every model downstream.
The SDK processes the microphone stream in the browser or the app and hands back a clean stream, so no extra server hop is added and raw audio does not leave the device.
Three model presets let you trade CPU for quality: speed for phones and busy machines, balanced and quality when you have headroom.
On the end user's device, in the browser via WebAssembly or in your native app. Your agent receives the already cleaned audio.
Yes. The Web SDK takes the microphone MediaStream and returns a processed one; publish its audio track to the LiveKit room instead of the raw microphone. The same works with any WebRTC stack.
Yes. The same models clean audio and video files on the device before you send them to speech-to-text.
Tell us about your product and platforms. Our team will get back to you and can help with the integration.
Ready to build? Open the Developer Portal and start a free trial.