Extract Audio from Video

A video file is a container holding separate streams: usually one video stream, one or more audio streams, and sometimes subtitles and chapter data alongside them. Pulling the audio out means discarding the video stream and keeping the audio, which sounds like it should be a pure copy. In practice it usually is not, because of what the audio is encoded as.

The audio inside an MP4 is almost always AAC, and inside a WebM it is Opus or Vorbis. None of those is MP3. Producing an MP3 therefore means decoding the original audio and encoding it again with a different codec, which is a second generation of lossy compression on top of whatever the video was already carrying.

Why that is usually fine anyway

The encoder here runs at a high variable-bitrate quality setting, averaging somewhere around 190 kbps depending on the material. Variable bitrate spends bits where the audio is complex and saves them where it is not, which is why it beats a fixed 192 kbps at a similar file size. At that setting the second generation of compression is inaudible for speech and very close to it for music, so a lecture recording, a podcast pulled out of a video, or a soundtrack you want on a phone all come through cleanly.

MP3 is the output because it is the format everything plays: car stereos, older phones, DAWs, fitness watches, hardware players that have never heard of Opus. If you would rather not add a generation of loss at all, the main audio converter on this site can produce WAV or FLAC instead, both of which store the decoded audio without compressing it further. They are substantially larger and worth it when the file is going into an editor rather than a playlist.

What determines how long it takes

Because the audio genuinely has to be decoded and re-encoded, extraction takes real processing time rather than being instant. A WebAssembly build of FFmpeg handles both halves of that inside a worker thread, which means your own processor sets the pace. Duration is what drives it: a three-minute clip finishes in seconds while a two-hour recording takes a while, and the resolution of the video barely matters since the video stream is discarded rather than decoded. Multi-track files, such as a film carrying several language tracks, yield the first audio stream in the container.