Convert MKV to MP3

A great many MKV files are not films. They are a lecture somebody filmed from the back of a room, a conference talk pulled off a stream, a church service, a band practice, a webinar, a podcast recorded with OBS because OBS records Matroska by default. The video adds nothing — a fixed shot of a lectern, a slide that has not changed in six minutes — and it makes the file twenty times larger than it needs to be.

That is what this conversion is for, and it explains why the result is so much smaller than people expect. An hour of talking as an MP3 is tens of megabytes rather than the gigabyte the MKV occupied, and the difference is almost entirely the picture nobody was watching.

This is the single most important thing on the page. MP3 as a format has room for exactly one audio stream. Matroska sets no ceiling worth naming, and files routinely use several — an original language and a dub, a mix and a commentary, or in an OBS recording a microphone track and a desktop track deliberately kept apart.

When there is more than one, the first track declared in the file becomes the MP3 and the others are discarded without a message. Note the wording: first in file order, not the track Matroska flags as the default. On material assembled by a ripping tool the two are often different, which is how somebody ends up with a director’s commentary where they expected the film. If your file has several tracks and you need a particular one, convert the whole thing to an MP4 or a MOV instead — those carry every track — and pull out the one you want in a player that lets you select it.

Some conversions stop before they start with the message that nothing in the file could be converted. That is not a damaged file and it is not a bug. It means the browser was asked whether it could decode the soundtrack and said no, and with the picture already discarded there was nothing left to write.

The codecs behind it are always the same short list: AC-3 and E-AC-3 from DVDs and Blu-rays, DTS in its several varieties, and Dolby TrueHD. Those are cinema audio codecs with their own licensing, and no browser ships a decoder for them — DTS and TrueHD are not even recognised as codecs by the reader. An MKV from OBS, a phone, a capture card or a download tool carries AAC or Opus and converts without trouble. A rip from a disc frequently does not.

Three bands: small file, balanced, high quality, defaulting to balanced. The band is expressed as a quality rather than a bitrate on purpose — the number that means "good" differs between codecs by more than a factor of two, and nobody converting a lecture knows whether they want 96 or 192 kbps.

For speech, the small band is genuinely fine and roughly halves the file against balanced, which matters when a term of lectures is going onto a phone. For music — a concert capture, a recorded performance — go to the high band, because the audio inside the MKV was already compressed once and this is a second pass over what the first one left behind. Two rounds of lossy compression on a cymbal is where it starts to be audible.

On the merits it is not the best choice. Opus and AAC both sound better than MP3 at the same size, and for speech the gap is large — an hour of talking as Opus can be a third of the MP3. If the destination is a phone you control, one of those is the better file.

MP3 wins on the thing that is not about sound at all. It has been readable by everything since 1993: a car stereo from 2006, a hospital PA, a school laptop, an old iPod, a cheap MP3 player bought for a walk, a corporate media system that has not been updated in a decade. When the file has to work somewhere you cannot test in advance, that is the whole argument, and it is usually the right one.

Where the Matroska file carries tags — title, artist, album, description, comment, date — they are read and written into the MP3 as an ID3 block at the front of the file. That is what a music app, a podcast player or a car display reads to show something other than the filename.

Most MKV recordings carry no tags at all, because recorders do not write them, so the usual result is an untagged MP3 named after the source file. If you are filing a series of talks, adding the tags afterwards in a tag editor is a two-minute job that makes the difference between a listenable library and a folder of unnamed files.

Much less than converting the video would, because the picture is discarded rather than re-encoded. Only the audio is decoded and compressed, and audio is a small fraction of the work in any video file. An hour-long recording is typically under a minute on an ordinary laptop.

The first MP3 you make in a session takes slightly longer to start. No browser can encode MP3 — it is the most requested audio format on the internet and the one WebCodecs uniformly refuses — so a WebAssembly build of the LAME encoder is downloaded once and then reused for everything afterwards. It is the only target in this bundle that needs one.

The free tier takes files up to 100 MB, and that is measured on the MKV going in rather than the MP3 coming out. A recording of a two-hour meeting can easily be a gigabyte of mostly-static video, which puts it over the line even though the audio inside it would have been forty megabytes.

It is an awkward limit for exactly this pair, and worth saying rather than hiding. Where a file is too large, the shortest path is a local tool that can copy the audio stream out without decoding anything — ffmpeg does it in seconds — or trimming the recording down to the part you actually want first. Everything else here runs on your own machine, so the ceiling is about how much work a browser tab should be asked to do rather than about bandwidth.