AI noise removal

Clean up your voice without uploading a thing

Remove hiss, hum, fans, traffic and room noise from speech with DeepFilterNet 3, an AI speech model that runs on your computer. Then even out the voice and set the loudness YouTube and TikTok play at.

Download free See Pro pricing

New in 1.1 Pro feature. Set it up and preview it free; exporting with it needs Pro.

Real exports from Captixa. Tap a style to preview it

At a glance

Voice cleanup in Captixa 1.1

Remove background noise, hum and room sound from speech with an AI model on your own computer. Voice boost and -14 LUFS loudness included.

remove background noise from videoai voice enhancerremove hum from audionormalize loudness -14 lufs
  • Noise modelDeepFilterNet 3, on your computer (8 MB)
  • StrengthLight ~12 dB · Medium ~22 dB · Strong
  • Remove hum50 Hz or 60 Hz, plus five harmonics
  • Loudness-14 LUFS, peaks under -1 dBTP
How it works

Voice cleanup in 4 steps

  1. 1

    Switch on Enhance voice

    In the Audio tab. The model downloads once (8 MB) and is checked before use.

  2. 2

    Pick a strength

    Light keeps a little room sound, Medium removes about 22 dB of noise, Strong removes as much as it can.

  3. 3

    Compare

    Flip Original and Enhanced, or press E, while it plays, at the same moment.

  4. 4

    Export what you hear

    The export takes its sound from the cleaned track, cuts included.

The details

What you get

Runs on your machine

The noise model runs on this computer. Your sound is never uploaded to be cleaned.

Fast

The whole track is cleaned once in the background with a progress bar: about 30 seconds for a 10-minute video.

Remove hum

Notches out 50 Hz or 60 Hz mains buzz and its first five harmonics, about 40 dB less hum, too narrowly to hear in the voice.

Voice boost

A low cut at 80 Hz, a little presence at 4 kHz, gentle 3:1 compression and de-essing, for a clearer, more even voice.

Loudness for every platform

Brings the track to -14 LUFS with peaks under -1 dBTP, set with one fixed gain under a limiter so speech does not pump.

Two waveforms

The timeline draws the original sound in grey and the cleaned sound in green, so you can see the noise removed between words.

Try it free, export with Pro

On the Free plan you can set up voice cleanup, preview it and hear it. When you export, Captixa lists the Pro features in use: upgrade, or export without them. Nothing in your project is switched off.

See Captixa Pro
FAQ

Questions, answered

Something else on your mind? Contact us and a real person will reply.

Does my audio get uploaded?

No. DeepFilterNet 3 runs on your computer. Only the model itself (8 MB) is downloaded, the first time you use it.

Can it help transcription?

Yes. Switch on "Clean before transcribing" in the Media tab and transcription runs on the cleaned sound, which helps on noisy recordings.

Will it put my audio out of sync?

No. Every stage keeps the timing to the sample, so the sound stays in sync with the picture and the captions.

Do I have to re-run the model to change the hum or loudness setting?

No. Hum, Voice boost and Loudness run after the noise model and are cached on their own, so changing one takes about a second.

Is it on the Free plan?

You can switch it on and hear the result on the Free plan. Exporting with cleaned sound needs Captixa Pro.

Try voice cleanup on your next video.

Download Captixa free, set it up and preview it. Export with Pro.

Download free See Pro pricing