Most recordings made outside a studio carry a layer of noise: the hiss of a cheap microphone, the hum of a fridge or air conditioner, a laptop fan, traffic rumble through a window. Listeners tune it out for a while, but over a long podcast or lecture it becomes tiring, and it gets worse every time the recording is made louder.
This tool reduces that kind of constant noise. Choose Light, Medium or Strong, optionally ask for a short before/after sample, and press Convert. You get a cleaned copy of each file and, if you asked for it, a second file that plays the first seconds as they were and then the same seconds cleaned, so you can judge the result before using it.
How noise reduction works and what it can remove
Noise reduction relies on a simple observation: steady noise sounds roughly the same all the time, while speech keeps changing. The denoiser splits the sound into many narrow frequency bands, estimates the noise level in each band, and turns down the parts of the signal that sit at or below that level, while louder, changing sound such as a voice passes through. Before that, a high-pass filter removes the deep rumble below about 60 to 90 Hz, where handling noise, traffic and much of the mains hum live but little of the human voice does.
This works well for hiss, hum, fans, air conditioning, engine drone and the general murmur of a quiet room. It cannot separate sounds that change as much as speech does: a dog barking, a door slamming, someone talking in the background or music playing behind the voice remain. Echo from a bare room is not noise in this sense either and stays largely as it is.
Recordings that benefit most
Podcasts and interviews recorded at home
Computer fans and room tone are constant and respond well. Medium usually removes most of it while keeping voices natural.
Lectures, meetings and voice memos
Recordings made on a phone or laptop from a distance tend to have audible hiss once they are boosted. Clean them before running Normalize Audio or Volume Booster.
Old tapes and digitised recordings
Cassette hiss and the steady noise of old equipment are exactly the kind of noise an FFT denoiser handles. Strong targets heavy hiss, at the cost of some brightness.
The three strengths and the before/after sample
Light cuts below 60 Hz and reduces noise by about 10 dB; it is gentle enough for music and for voices with little noise. Medium, the default, cuts below 80 Hz and reduces by about 20 dB, the best balance for most speech. Strong cuts below 90 Hz, reduces by about 30 dB and adds a second denoiser, a non-local means filter that removes the remaining hiss; it is the slowest and can make voices sound thinner or slightly processed, so compare before you commit to it.
Switching on Add a before/after sample produces an extra file ending in -before-after: the first 8 seconds of your recording exactly as they were, followed by the same 8 seconds after cleaning. Choose a recording that has noise in its first seconds, or trim a representative stretch first. The cleaned file ends in -clean. Output format keeps the input's format by default, with MP3, WAV, M4A and FLAC as alternatives; WebM recordings become MP3. Processing uses ffmpeg's FFT denoiser running in your browser through the site's audio converter, a one-time download of about 31 MB.
Как пользоваться инструментом
- Drop one or more MP3, WAV, OGG, M4A, AAC, FLAC or WebM files on the box above, or click it to browse.
- Choose a Strength: Light, Medium or Strong.
- Switch on Add a before/after sample if you want to compare, and choose an Output format.
- Press Convert, listen to the before/after file, and download the cleaned recording.
Limitations
- Sudden or changing sounds, such as barking, voices in the background, music and clicks, are not removed.
- Room echo and reverb are not reduced.
- The stronger the setting, the more the voice itself can be affected, sounding duller, thinner or slightly watery.
- The high-pass filter removes deep bass, so use Light, or skip noise reduction, for music with important low notes.