MP4 to MP3

An MP4 is a container holding separate video and audio tracks. Getting an MP3 out of it means finding the audio track, decoding it, and re-encoding it as MP3 — the video is discarded before anything touches it.

That has a useful consequence. Browsers cannot decode HEVC, which is what an iPhone records in by default, so a browser often cannot play an iPhone video. It can still extract its audio perfectly, because the video codec is never invoked. A conversion that decoded the whole file would fail on exactly the files people most often have.

Browsers decode MP3 but do not encode it — there is no MP3 encoder in WebCodecs. The encoding here is LAME, the reference open-source MP3 encoder, compiled to WebAssembly and loaded only when you actually convert to MP3.

It is the same encoder desktop software has used for twenty years, producing standard files that play everywhere. Because it is fetched on demand, pages that convert to WAV or M4A never download it.

The source was almost certainly already lossy, and encoding to MP3 discards a little more. The effect is called generation loss and it compounds with every pass.

At the High setting, one generation is inaudible on ordinary material. It becomes audible when the source was already poor — a low-bitrate stream, a voice recording from a phone — or when the file has been through several conversions already.

Where you have a choice, convert once from the best source you have rather than from something already converted.

Quality sets the bitrate. High is the default and is transparent for music. Low produces a much smaller file and is fine for speech.

Mono is the right choice for a lecture, an interview or a podcast where both channels carry the same thing. On its own it does not shrink the file — the encoder holds the bitrate — but it lets you drop the quality much further before anything is audible, and that is where the saving comes from.

Sample rate is best left alone unless the file is speech, where 22 kHz saves space with no perceptible loss.

Trim takes a section without a separate editing step — useful for pulling one song out of a long recording.

The decoding here is the browser's own, which is fast and needs no download — and which has real gaps.

WMA (Windows Media Audio), AC-3 and DTS (cinema surround tracks), AMR (old phone voice recordings) and RealAudio are not decodable by any browser. If your file uses one you are told which, rather than being left with a failure that says nothing.

HEVC video cannot be decoded either — but that does not matter here. Extracting audio discards the video track before anything tries to decode it, so an iPhone recording converts perfectly well even though the browser could not play its picture.