What noise reduction can and cannot do
Background noise comes in several kinds. Some is steady, such as the hiss of a microphone, the hum of a fan, or the buzz of electrical interference. Some changes constantly, like traffic, other people talking, or clattering dishes. Software finds it much easier to deal with the steady kind.
This tool uses ffmpeg's FFT-based denoise filter (called afftdn). It analyses the audio in the frequency domain, estimates the noise floor, and lowers the parts of the signal that look like that floor. It works best on constant noise and is much less effective on variable noise.
When to use it
It is useful for tidying a voice memo, interview, or lecture that has an even background hiss or hum before you share it or keep working on it. A mild cleanup can make speech easier to follow.
It will not turn a bad recording into a studio one. Noise that overlaps with the voice in the same frequencies cannot be removed without affecting the voice, so heavy processing can leave speech sounding thin, watery, or slightly robotic. Treat it as a light cleanup step. If a recording is also too quiet, you could reduce noise first and then raise the level with the Volume Booster, since boosting first would make the noise louder too.
How to reduce noise with this tool
- Drop an audio file onto the upload area, or click it to browse.
- The file name and size appear. There are no sliders: the filter runs with one fixed noise floor setting (-25), so the tool has nothing to adjust.
- Click Run. The first run downloads the ffmpeg engine (roughly 30MB) and shows a progress bar.
- Play the processed result in the audio player and compare it with your original.
- Click Download. The output is an MP3 named after your original file plus "-denoise", for example
memo-denoise.mp3. - Start over clears the file so you can try another.
The audio is processed in your browser with ffmpeg.wasm and is not uploaded.
Setting expectations and avoiding disappointment
Because there is no strength control, you get one level of reduction. If the result is not what you wanted, there is no gentler or stronger setting to try. A recording with only light hiss may improve noticeably, while one with loud or changing noise may change very little.
Listen on headphones and compare against the original, paying attention to the voice itself. If it sounds worse than the noise did, the original may be the better file for your purpose.
The output is always MP3, so the audio is re-encoded and a small amount of quality is lost. Keep the original file, and always run the tool on the original rather than on a file you have already processed. If you only need part of a recording, trim it with the Audio Cutter first so there is less to process.