Tools›Audio›AI Voice Isolator

PIXELOKAY AUDIO STUDIO · FREE & BROWSER-BASED

AI Voice Isolator

Turn down background noise and bring dialogue forward. Fine-tune noisy interviews, lectures, voice notes and podcasts right on your device.

✦ Neural & classic filters✦ Before / after preview✦ No audio uploads✦ WAV export

01 / CLEAN YOUR AUDIO

Voice restoration workspace

Choose a recording, shape the cleanup and hear the difference.

Your audio stays local

AUDIO PREVIEW

Listen to the difference

Waiting for audio

♫

Drop an audio recording here.

Or choose a file to begin cleaning your recording.

MP3, WAV, M4A, AAC, OGG, WebM and other browser-supported formats · max 100 MiB

Download enhanced audio

Uncompressed WAV; choose PCM precision. Source files are never modified.

ENHANCEMENT CONTROLS

Shape the cleanup

Choose an engine 1 OF 2

Studio filterBuilt-in · works completely offlineNeural AI (RNNoise)Free model · first use requires internet

Neural mode is opt-in. RNNoise weights ship with a free JS package downloaded after you choose Download & Load. Your audio stays in the browser. Studio mode needs no model.

Starting points PRESETS

Voice cleanup FINE TUNE

Noise detection ADVANCED

How to learn background noise

Choose first 2 seconds only if that section mostly contains background noise. Noise profiling is used by Studio filter; neural mode uses its own learned model.

Recording format OUTPUT

Preserve stereo channels Prevent sample clipping (peak safety)

For mono output, stereo inputs are downmixed before processing. One or two channels are supported.

Select audio to begin

02 / WHY PIXELOKAY

Bring speech forward, without a studio.

◉

Hear what changed

Compare original and processed waveforms, then play each version before downloading.

✦

Choose your approach

Use our offline spectral filter or a genuinely neural, free RNNoise engine when online.

⌁

Privacy by design

Files are decoded and processed on your own device. No account or paid API required.

Frequently asked questions

Is AI Voice Isolator actually an AI tool?

Its optional Neural AI mode uses an open-source RNNoise speech-enhancement model downloaded on demand. The default Studio filter is traditional statistical spectral denoising, not AI. Both run in your browser.

Can it completely separate voices from music or another person?

No. This is a speech noise suppressor, not a perfect source-separation system. Background music, overlapping speakers, severe clipping, and strong reverberation may remain or cause artifacts. Preview before downloading.

Do you upload my recordings?

No. This implementation never sends the audio to a server. Optional neural mode downloads program code from an external CDN on first use; the CDN can see standard network metadata, not your recording.

Which audio files work?

The browser decodes its supported audio formats, usually WAV, MP3, M4A, OGG and WebM (varies by browser). Enhanced output downloads as PCM WAV at the decoded sample rate. Up to 100 MiB compressed files and 24 million decoded samples are supported.

Will exporting change the original file?

No. Export creates a new file from the enhanced selection. The original stays untouched. Changes to controls require pressing Apply again before the updated result can be downloaded.

Using AI Voice Isolator

Import a supported recording, choose a selection and adjust Studio cleanup or optional Neural RNNoise. Apply, audition the original/result/removed sound and export WAV. Keep the source recording separately.

Studio mode is conventional DSP; only optional RNNoise is neural and requires an external runtime download. No microphone capture, batch, MP3 export, perfect music separation or dereverberation is supplied. Decoding depends on browser support; file/channel/decoded-sample budgets apply. Peak safety measures sample peak rather than true peak. Reconstruction and noise-removal quality are not guaranteed.

Clean speech with conventional DSP or optional neural denoising

Import supported audio and begin with the Studio spectral cleanup controls. Optional RNNoise neural denoising is a separate mode that loads an external runtime when requested. Compare the result before exporting PCM WAV.

Example workflow

Import a short recording and listen to the source before changing processing settings.

Adjust conventional cleanup and compare it with the original; try neural denoising only if you accept its external runtime dependency.

Review the processed audio and export 16-bit or 24-bit PCM WAV if suitable.

Authored example: Use a voice memo with steady background hiss. Compare a modest spectral reduction with optional neural denoising and listen for speech artifacts; do not assume either setting separates overlapping speakers.

Limits and model handling: Neural Apply downloads a runtime from an external provider; the application code does not intentionally upload the source recording for inference. External hosts receive network requests. Cleanup does not guarantee source separation or dereverberation. There is no microphone capture, batch workflow or MP3 export, and peak safety is sample-peak rather than true-peak limiting.

AI Voice Isolator questions

Does the Studio mode require an AI model?

No. Conventional spectral processing is independent of the optional RNNoise path.

Does neural denoising separate voices from music?

It is not guaranteed source separation. Review the output for artifacts and use a dedicated separation workflow when that is required.