FIELD GUIDE
Clean up a voice recording without losing the words
Noise reduction works best when the source is understandable and the unwanted sound is fairly steady. Use this workflow to compare the original with the processed recording and decide whether the result is clear enough for its next use.
Know what noise reduction can and cannot do
The tool runs a speech-enhancement model on the recording and returns a cleaned audio file. It is intended to reduce background noise around speech, not to remove every sound or reconstruct information that was never captured. Room hiss and steady ambience are different from overlapping speech, music, clipping or a microphone rubbing against clothing.
If several people speak at once, the model may leave parts of another voice or soften the speaker you want to keep. Treat the result as a listening aid, not as proof that every background sound has disappeared.
Start with the clearest source you have
Choose the original recording instead of a copy that has already been compressed or processed. A file with clear speech gives the model more useful detail to work with. If the voice is very quiet, clipped or buried beneath loud music, try to locate another recording before expecting software to repair it.
The tool accepts browser-readable MP3, WAV and M4A audio. A file extension does not guarantee the codec inside is supported. If a file fails to load, try exporting a common PCM WAV or AAC M4A copy from the recording app.
Compare matching moments, not different passages
Play the original and cleaned versions at the same point in the recording. Listen once for the words, then again for quieter consonants, breaths and room tone. Headphones make it easier to notice if a syllable has become dull or a short pause sounds unnaturally empty.
Use the waveform to return to the same passage. Do not compare a quiet part in one version with a loud part in the other; level and content differences can make the processed file seem better or worse than it is.
Decide whether the cleaned voice is still natural
A useful result keeps the words easy to understand without making the speaker sound watery, metallic or cut off. Check names, low-volume phrases and the beginning and end of sentences. If speech quality matters more than a quiet background, keep the original whenever processing changes the voice too much.
Understand the downloaded file
The result is exported as M4A with AAC audio. The tool processes the recording on the device and does not upload it for processing. Keep the browser tab open while a file is being processed, then play the downloaded file in the player you plan to use.
Troubleshoot by changing the source first
If the result is distorted, compare with the original again and try a cleaner source or a shorter representative passage. If playback works but downloading fails, check the browser’s media and storage support. Noise reduction cannot separate a chosen speaker, restore clipped syllables or guarantee broadcast-quality audio.
Before you use the recording
Confirm that the intended words remain audible, that no important short phrase was softened, and that the saved M4A opens correctly. The processed recording is a new export; retain the original so you can return to it if the cleaned version changes speech detail.