Extract the audio from a video without re-encoding it

A soundtrack is a separate stream

An MP4, MOV, MKV or WebM file is a container: a box holding separate streams side by side, each with its own timestamps. The picture is one stream; the sound is another, already compressed with an audio codec — AAC in almost every MP4 and phone recording, Opus in most WebM files, sometimes MP3, FLAC, AC-3 or plain PCM. Extracting the audio does not require "converting" anything. The sound is already a finished, compressed stream sitting next to the video, and it can be lifted out whole.

Copying versus re-encoding

There are two ways to get that stream into a file of its own.

• A stream copy takes the compressed audio packets and writes them into a new container unchanged. Nothing is decoded, so the output is bit-identical to what was inside the video — the definition of "without quality loss". It is also fast: the work is reading and writing a few megabytes, not processing sound.
• A re-encode decodes the audio to raw samples and compresses it again with another codec. AAC and Opus are lossy codecs, so the original encoding already discarded detail the ear is unlikely to miss; compressing the result again discards a second round. A single lossy-to-lossy step at a sensible bitrate is usually hard to hear, but it can never be better than the copy, and it is slower.

That is why Original is the default and the recommendation. Re-encoding makes sense only when something downstream needs a specific codec: an old car stereo that only reads MP3, an editor that wants WAV, a voice-note app that wants Opus.

What "Original" gives you, codec by codec

A copied stream still needs a container, and the right one depends on the codec. Getting this wrong is the classic ffmpeg mistake — -c:a copy output.mp3 on an AAC track fails, because an MP3 file cannot hold AAC. This page reads the codec first and picks the matching container:

The format button for Original shows the extension you will actually get before you run it, so there is no surprise .mka in your downloads.

WAV is bigger, not better

WAV stores uncompressed PCM samples, so its size is plain multiplication: sample rate × channels × bytes per sample. For 16-bit stereo at 48 kHz, the rate most video uses, that is 48,000 × 2 × 2 = 192,000 bytes a second. A minute is 11.52 MB; an hour is 691.2 MB. The same stereo track as AAC at a typical 128 kbps is 128,000 ÷ 8 = 16,000 bytes a second — twelve times smaller.

Decoding to WAV loses nothing further, but it cannot restore what the AAC encoder removed when the video was made. What WAV buys you is convenience: every audio editor, DAW and transcription tool opens it, and cutting or processing it causes no additional generation loss. For keeping the sound as it was, Original is identical and a fraction of the size.

WAV also has a hard ceiling. Its size fields are 32-bit, so the sample data cannot pass 4 GiB — 4,294,967,295 bytes. At 192,000 bytes a second that is about 22,370 seconds, roughly 6 hours 12 minutes of 48 kHz stereo; a 5.1 track at the same rate, at 576,000 bytes a second, runs out after about 2 hours 4 minutes. The format picker checks this against your file and section before you start.

Why MP3 is the hard one in a browser

This page encodes with WebCodecs, the audio and video encoders the browser itself provides. Chrome, Edge and Firefox ship Opus encoders, and Chrome and Edge ship AAC on most desktop systems; Firefox has no AAC encoder. None ships an MP3 encoder, because nothing on the web platform ever needed to produce MP3; browsers only ever had to play it. Encoding MP3 in a tab means shipping an encoder as WebAssembly or JavaScript alongside the page.

That is exactly what this page does. When the audio inside the video is already MP3, Original and MP3 are the same lossless copy. For anything else, choosing MP3 loads LAME — the reference MP3 encoder — compiled to WebAssembly, on demand, so nobody who never picks MP3 downloads it. The encoding still happens on your machine. Before reaching for MP3, consider M4A: almost every device made in the last decade that plays MP3 also plays AAC, and AAC generally sounds better at the same bitrate.

The same jobs in FFmpeg

First find out what you have, because the copy command depends on it:

ffprobe -v error -select_streams a:0 -show_entries stream=codec_name -of csv=p=0 input.mp4

Then:

-vn leaves the video out; -c:a copy moves the audio untouched. If the file has more than one audio track — a film with several languages, a recording with a separate microphone — add -map 0:a:1 to pick the second. This page takes the primary track.

When there is no audio at all

Some videos have no sound stream: screen recordings made without a microphone, clips exported muted, animated GIFs converted to MP4. There is nothing to copy or decode, and the page says so plainly instead of producing an empty file. The ffprobe command above prints nothing for such a file, which is the quickest way to check before you go looking for a soundtrack that isn’t there.