Free tools · Run in your browser
Free online audio tools
Short answer
Four audio tools, free and with no upload: trim an audio file, normalize its loudness, remove silence and remove background noise with an AI speech model. They run in your browser; on a 30 second speech recording they took 0.01 to 0.87 s for trimming, normalizing and silence removal and 0.50 to 0.60 s for noise removal.
Turn the speech into words
Edit and clean
Convert and extract
Key takeaways
- Everything happens in your browser: the file is not uploaded and there is no account.
- Trimming an MP3 does not re-encode it: the bytes are copied, which is why a cut of our recording took 0.01 to 0.10 s.
- Normalizing moved our speech recording from -23 to -16 LUFS; the page shows the loudness and peak before and after.
- The noise remover raised the signal-to-distortion ratio on every test file, by up to 6.0 dB at the Gentle setting; FFmpeg's default denoiser changed the same files by a fraction of a decibel.
- The audio converters turn between MP3, WAV, AAC, M4A, FLAC and OGG, and extract the audio from a video.
Measured: time and size
The video and audio tools run a single-thread WebAssembly build of ffmpeg inside your browser tab. The first tool you use downloads that engine, about 30 MB, once per visit. Timings below were measured on a 10 second 1920 × 1080 H.264 clip (13.5 MB) or a 30 second speech recording, in a headless browser on a server, so a laptop will be quicker and a phone slower.
| Tool | What it does | Time | Output |
|---|---|---|---|
| Trim | Cuts a start to end range | 0.01 to 0.10 s | 129 KB to 161 KB |
| Normalize | Sets the loudness to a target | 0.81 to 0.87 s | 405 KB to 420 KB |
| Remove silence | Cuts dead air | 0.19 to 0.24 s | 296 KB to 357 KB |
| Remove noise (Gentle) | AI speech model | 0.50 to 0.60 s | 30 seconds of speech |
The ranges cover the settings measured on each tool page.
Measured: loudness normalization
The normalize tool sets the integrated loudness to -16 LUFS with a true-peak ceiling of -1.5 dBTP. These are the recordings we measured, before and after.
| Recording | Before (LUFS, LRA, peak) | After (LUFS, LRA, peak) | Time |
|---|---|---|---|
| speech as recorded | -23, 2.5, -6.1 | -16, 2.7, -1.4 | 0.83 s |
| very quiet speech (20 dB lower) | -43, 2.5, -26.1 | -16, 2.7, -1.4 | 0.81 s |
| loud speech (10 dB higher, limited) | -13.3, 2.9, 0 | -15.9, 2.8, -2.2 | 0.87 s |
Measured: removing silence
| Recording | Length before | Length after | Removed |
|---|---|---|---|
| speech with natural pauses | 30 s | 20.611 s | 9.389 s (31.3%) |
| the same speech with fan noise at 15 dB | 30 s | 29.44 s | 0.56 s (1.87%) |
| the same speech with fan noise at 5 dB | 30 s | 29.806 s | 0.194 s (0.65%) |
Measured: removing noise
A public domain speech recording was mixed with a real fan recording and a real street recording at four noise levels. Gain is the change in signal-to-distortion ratio against the clean speech; higher is cleaner.
| Noise | Noise level vs voice | AI model, Gentle | afftdn defaults |
|---|---|---|---|
| Studio fan | 0 dB | +6.0 dB | +0.00 dB |
| Studio fan | 5 dB | +4.8 dB | +0.02 dB |
| Studio fan | 10 dB | +3.1 dB | +0.09 dB |
| Studio fan | 15 dB | +0.8 dB | +0.23 dB |
| Residential street | 0 dB | +5.6 dB | +0.00 dB |
| Residential street | 5 dB | +4.7 dB | +0.01 dB |
| Residential street | 10 dB | +3.5 dB | +0.02 dB |
| Residential street | 15 dB | +1.6 dB | +0.03 dB |
The model is built for speech. The comparison column shows why the default denoiser is not enough on its own: it moved the same files by a fraction of a decibel.
Which tool for which job
| You want to | Use | Why |
|---|---|---|
| Cut an intro or an outro | Trim | Start and end in seconds; an MP3 is not re-encoded |
| Make episodes the same loudness | Normalize | Sets the loudness to a target and shows the measurement |
| Tighten a spoken recording | Remove silence | Cuts long pauses automatically |
| Clean a noisy voice | Remove noise | An AI speech model, tested on real fan and street noise |
| Change the format | Audio converters | MP3, WAV, AAC, M4A, FLAC, OGG |
Limits to know about
- Files up to 100 MB.
- The engine runs in your tab, so a very long recording is slower than in a desktop app.
- Noise removal is built for speech; music is not its target.
Questions people ask
Are the audio tools free?
Yes, with no account and no watermark.
Is my audio uploaded?
No. The tools run in your browser and the file stays on your device.
Does trimming reduce quality?
Not for MP3: the tool copies the data between your start and end, so the sound is identical to the original.
Does noise removal work on music?
It is designed for speech. The page shows measurements on speech only.
Which formats can I convert between?
MP3, WAV, AAC, M4A, FLAC and OGG, among others, on the audio converter pages.
Sources and how we checked
- FFmpeg Documentation: ffmpeg (read 2026-10-09)
- EBU R 128: the European loudness recommendation for broadcast audio (read 2026-10-06)
Reviewed October 9, 2026. Every figure on this page comes from the measurements behind the linked tool pages, which are saved in data/seo/hub/evidence/2026-10-09.