What noise reduction can and cannot fix
Noise reduction works best on steady, unwanted sounds such as a fan, air conditioner, computer hum, or consistent room hiss. It can make speech easier to understand, but it cannot reliably reconstruct words covered by a loud interruption. Clipping, heavy echo, wind blasts, and music mixed into speech may need a more specialized editor or a better recording.
The audio denoiser is designed for voice-oriented cleanup and processes supported audio locally in the browser. Keeping a recording on the device is useful when the file contains an interview, meeting, private voice note, or unreleased podcast.
Start with the cleanest source
Noise reduction cannot improve information that was never captured. Record close to the microphone, reduce room reflections, and keep the input level below clipping. A short pause containing only room noise can help you judge what the unwanted sound is like, but do not assume every recording has a perfectly isolated noise sample.
Keep an untouched copy before processing. This makes it possible to compare the cleaned result and return to the source if the filter removes consonants or makes the voice sound metallic.
Use a light pass and listen for artifacts
The goal is not silence between every word. Natural room tone can sound more comfortable than a recording that repeatedly drops to digital silence. Apply enough reduction to lower the distraction, then listen with headphones and on ordinary speakers.
Watch for:
- watery or metallic speech;
- missing “s”, “t”, or breath sounds;
- pumping as the noise level changes;
- abrupt silence between words;
- a voice that sounds unnaturally thin.
If you hear these artifacts, reduce the strength or start from a better source. A slightly noisy but natural voice is often more useful than an aggressively processed voice that is difficult to trust.
A complete voice-cleanup workflow
- Trim audio to remove obvious setup and handling noise at the edges.
- Use noise reduction with a moderate setting.
- Preview a quiet section, a normal sentence, and the loudest section.
- Normalize the audio only after cleanup if the overall level needs balancing.
- Remove long, unnecessary pauses with silence removal, leaving natural breathing room.
- Export a lossless working copy, then convert or compress a separate delivery version.
Do not normalize first if the noise is very quiet relative to the voice; raising the whole file can make the background more noticeable. Final loudness should be checked after the main cleanup decisions are finished.
Privacy and consent
Voice recordings can reveal private conversations and identifying information. Review where a tool processes the file before using it. A local browser workflow means the source is read and processed on the current device, but you should still protect exported files and obtain permission before editing or sharing someone else's recording.