The audio tools here run on the Web Audio API, which decodes whatever you give it into raw 32-bit floating-point PCM samples before doing anything else. That is why editing feels instant once the file has loaded — every operation after decode is arithmetic on an array in memory, with no seeking or container parsing — and it is also why memory, rather than file size, is the real constraint. Five minutes of 44.1 kHz stereo decodes to roughly 50 MB of floats no matter whether it arrived as a 4 MB MP3 or a 50 MB WAV.
That decode step also determines what is lossless and what is not. Trimming, joining and normalising all operate on the decoded samples, so exporting to WAV afterwards preserves exactly what you heard in the preview. Exporting to MP3 re-encodes, which means a fresh generation of lossy compression on top of whatever the source already had. If you are going to run several edits in sequence, keep WAV between the steps and only encode to MP3 at the very end.
Loudness is the part people most often get wrong. Peak normalisation raises the file until its single loudest sample hits your target — which does nothing for perceived volume if that peak is one stray click. Loudness normalisation measures the average across the whole programme and is what actually makes something sound louder. The widely used targets are around −16 LUFS for stereo podcasts and −19 LUFS for mono, while the major streaming platforms normalise playback to somewhere near −14 LUFS, which is why mastering a track louder than that gains you nothing on Spotify.
Video is the one area where the browser is genuinely at a disadvantage. Decoding is hardware-accelerated on virtually every device; encoding often is not, so compressing video runs on the CPU and can take longer than the clip itself. For a 30-second screen recording that is fine. For a 40-minute 4K file, a desktop encoder will still finish first — that is an honest limitation, not something a better web tool fixes.
One more expectation to set: vocal removal works by cancelling what is identical in the left and right channels, which is where a centre-panned lead vocal lives. On older mixes with a dry centred vocal it can be startlingly effective. On modern productions with stereo-widened, heavily reverbed vocals, it removes some of the vocal and a fair amount of the bass and snare along with it. It is a useful trick, not stem separation.
Choosing between the audio editors
Several of these accept the same files. What differs is the operation:
| Tool | Operation | Lossless to WAV? |
|---|---|---|
| MP3 Cutter & Audio Trimmer | Cut a region, apply fades | Yes |
| Audio Joiner | Concatenate with gaps or crossfade | Yes |
| Audio Normalizer | Raise level to a peak or LUFS target | Yes |
| Silence Remover | Detect and delete quiet sections | Yes |
| Audio Speed & Pitch Changer | Time-stretch or pitch-shift | Resampled |
| Audio Converter | Change container and codec | Only if target is WAV |
| Vocal Remover | Mid/side cancellation | Alters the mix |