Extract Audio From Video

Which option actually keeps the quality

The sound in a video is already compressed — usually AAC, sometimes AC-3, Opus or MP3 — and the choice on this page is really between copying that out and encoding it again. Keeping the original does the first: the encoded audio is lifted into a new file without being decoded at all, so it is bit for bit what the video contained, and it takes seconds however long the film is. Nothing else here can be better than that, because nothing else can be as good as the source.

Choosing MP3 or WAV instead means a conversion, and each has a specific reason to exist. MP3 is a second generation of lossy compression on top of the first — two codecs that disagree about what is inaudible, so the result is slightly worse than either would be from a clean source. It is the right choice anyway when the file has to play in a car, on an old device or in software that refuses anything else. WAV decodes the sound and stores it uncompressed: it adds no further loss and cannot undo the first one either, so it is for editing rather than for listening, and it is very large — around 10 MB a minute in stereo.

Two limits come straight from the formats. MP3 and WAV are written in stereo here, so a film with 5.1 surround is folded down — audible on the dialogue of a cinematic mix, where the centre channel carries the voices and folding changes its balance against everything else. And keeping the original needs a container that suits the codec, which is why the option is unavailable for a few very old formats: there is nowhere valid to put the stream, so converting is the only route.

None of this improves on what the video had. A video encoded with 128 kbps audio yields 128 kbps audio, and asking for 320 kbps MP3 produces a larger file containing the same sound plus the new encoder's own guesses. The bitrate the page reports for the video's own track is the number worth matching rather than beating.

When you want something else

ffmpeg copies the audio out with no conversion at all and picks the right container automatically: ffmpeg -i in.mp4 -vn -c:a copy out.m4a. -vn drops the video, -c:a copy guarantees the sound is untouched. For a folder of files, a loop over that line is the whole job, and there is no memory ceiling.

To keep surround intact rather than folding it to stereo, the sound has to stay in a format that can hold it — copying the original AC-3 or AAC track out does exactly that, where MP3 and WAV here will not. And if the audio is destined for editing, take it out losslessly and convert inside the editor, so there is only ever one generation of loss rather than two.