Clearvoice
Checking WebGPU

AUDIO RESTORATION

Let the voice through

Clean background noise and enhance speech fidelity directly on your device.

Stays with youYour recordings never leave this browser.

01 RECORDING

02 LISTEN & COMPARE

Your result will appear here
Original 0:00 /0:00
Clean 0:00 /0:00

Videos are converted to audio automatically. Exports are mono WAV files.

Make room for speech

NVIDIA NemotronLabsRE-USE · Speech enhancement
LOCAL

For conversations, interviews, voice notes, and recordings that need a clearer voice.

A good balance for everyday speech recordings.

Long recordings are joined with overlapping sections.

100%
More originalMore cleanup

Blend the clean result with the original. Adjust after processing without running the model again.

The model downloads once (38.6 MB), then runs locally.

BEHIND THE CLEANUP

Your browser is the studio.

Clearvoice runs NVIDIA's original 9.6-million-parameter NemotronLabs RE-USE model on your graphics processor through WebGPU. There is no processing server and no local helper to install.

Select an audio or video file. Your browser extracts and converts the audio, the model cleans overlapping sections, and you can listen and download the result as a mono WAV.

This model is designed for speech. Music, laughter, and secondary speakers may be reduced. A sample-rate setting changes the output bandwidth; it cannot guarantee recovery of missing detail.

Use current Chrome or Edge with hardware acceleration enabled. Media decoding depends on the browser's supported codecs. Files are limited to 250 MB and 20 minutes to keep browser memory manageable. Full-bandwidth mode uses shorter sections.

The model is cached on this device after its first download. Recordings are held in memory and are discarded when the page closes.

Original model & documentation ↗Model license: OpenMDW-1.1 ↗