Does it remove reverb or separate music?

No. It targets steady noise using a measured profile, not room echo, competing voices or music separation. It is not a neural voice replacement.

Why does the noise sample matter?

Including speech in the profile can remove voice frequencies. Select at least 0.15 seconds with only background noise and review the result.

What are the limits and export format?

Mono/stereo audio up to 50 MB, 5 minutes, 96 kHz and 30 million decoded channel samples. Output uses the displayed browser-decoded rate and channels as 16-bit WAV. Processing and previews remain local.

Speech Cleanup Studio

Learn a noise-only sample, reduce steady background noise, compare synchronized before/after playback and export WAV.

Noise profile

Choose at least 0.15 seconds containing only background noise. Speech in this sample can cause the voice to be removed.

Start (seconds)End (seconds)

Reduction strength: 1.5Processed mix: 100%

Spectral noise reduction for steady hiss, fans and room noise. This is not an AI voice replacement, reverb remover or voice/music separator. Gentle settings usually preserve speech better.

Listening desk

No recording

Original

Open a recording to identify a short noise-only section.

Cleaned

Your processed recording appears here.

Export uses the displayed browser-decoded sample rate and channel count as 16-bit PCM WAV. Peaks exceeding the output ceiling are scaled down to avoid clipping.

Local processing · 50 MB / 5 minutes · no upload or model download

Noise reduction can create metallic artifacts and cannot repair clipped speech. Keep the original recording, review the entire output, and reduce strength if speech sounds unnatural.

Loading Tool…

What Is the Speech Cleanup Studio?

A local, profile-based spectral noise reducer for steady room noise, hiss and fans. It measures a selected noise-only passage, applies spectral subtraction with overlap-add reconstruction, and lets you adjust reduction strength and processed mix.

How It Works

Open a recording, listen for a noise-only passage and enter its start/end times or use the playhead. Run reduction, compare the same passage with A/B playback, then download a 16-bit PCM WAV. Adjust strength and rerun if speech sounds metallic.

When to Use It

Clean a voice memo, interview or narration recorded with a consistent background noise. This complements compression and transcription rather than duplicating them.

Frequently Asked Questions

Does it remove reverb or separate music?: No. It targets steady noise using a measured profile, not room echo, competing voices or music separation. It is not a neural voice replacement.
Why does the noise sample matter?: Including speech in the profile can remove voice frequencies. Select at least 0.15 seconds with only background noise and review the result.
What are the limits and export format?: Mono/stereo audio up to 50 MB, 5 minutes, 96 kHz and 30 million decoded channel samples. Output uses the displayed browser-decoded rate and channels as 16-bit WAV. Processing and previews remain local.

Found an issue? Send Feedback. See our data sources and update policy for tools that depend on changing information.