Convert MP4 to AAC

The output is ADTS: a run of AAC frames, each carrying a short header describing its own sample rate and channel count, with nothing wrapped around them. It is the shape AAC takes when it is being transmitted rather than stored — the form a broadcast chain, a segmenter or a decoder on a small device expects to be handed.

An MP4 holds the same AAC audio inside a full container, with an index describing where every frame sits and a header stating the duration. This conversion strips that structure away and leaves the frames. Nothing about the audio changes shape; everything about how software finds its way around it does.

The words point the wrong way. AAC is a codec and M4A is a file, and someone who wants "the AAC out of my video" to put on a phone is describing an M4A. That file contains exactly the same AAC audio with a container around it, plays everywhere, and carries a title and cover art.

The bare stream is the right answer only when something downstream demands it. If you cannot name the thing that demands it, take the M4A and stop reading — you will get a file that works everywhere instead of one that half your software refuses to open.

No metadata of any kind. There is nowhere in an ADTS stream to record a title, an artist, an album, a track number or a piece of artwork, so a library that imports one shows a file name and nothing else. There is also no chapter list and no way to add one.

And no seek index. A player wanting to jump to the twentieth minute of a containerised file reads an offset from a table; with a raw stream it has to estimate from the bitrate and scan. That is why durations on ADTS files sometimes read wrong until the file has been played through, and why scrubbing feels imprecise.

An ADTS stream carries one audio track and that is a hard property of the format rather than a limitation of this converter. A video holding an original language, a dub and a commentary produces a stream containing the first of them, and the others are gone.

If the source has more than one and you need them, the audio has to go into something with room: M4A and Ogg both hold several. Check what is in the file before assuming the right track came out, because discovering the wrong language an hour into a build is an expensive way to learn it.

Both the MP4 and the ADTS stream hold AAC, so a straight copy of the frames would have been possible. This conversion decodes and encodes instead, which costs one generation of quality — small on speech, measurable on music, and worth knowing about before a file becomes a master for something else.

For a delivery stream that is usually irrelevant: the audio is on its way to being compressed for transmission anyway, and one more pass at the balanced band changes nothing a listener will hear. For anything being kept, take the Extract audio tool on this site with the M4A option, which copies the original packets without a decoder running.

Segmenting for HLS, where the audio is about to be sliced into short pieces and the container would only be in the way. Embedded and automotive players that read ADTS directly because implementing a full MPEG-4 parser on the device was never worth it. Test material for a decoder, where you want frames and no structure around them.

The common thread is that something else is going to impose the structure later. A raw stream is an intermediate, not a destination, and treating it as a destination is how you end up with a folder of files nothing will index.

Three bands, defaulting to balanced, resolved by the encoder into a bitrate appropriate to AAC. Balanced is right for speech and for most delivery work; the high band is worth it when the stream is music that will be listened to attentively rather than played under something.

The small band exists for constrained links — a device on a slow connection, or a feed where the audio budget is fixed and small. AAC holds up better than most codecs when squeezed, which is part of why it became the default for streaming in the first place.

VLC and FFmpeg read raw ADTS without hesitation, and so do the command-line tools that anyone building a delivery chain already has. Browsers, phones and most desktop music players do not — they identify audio by container and see something they have no parser for.

That is worth setting expectations about, because a file that will not open looks broken and is not. If somebody else has to play the result, send them an M4A. If a machine has to read it, the stream is correct.

The work happens in the browser tab on your own processor and nothing carrying the video is sent anywhere. The free tier takes files up to 100 MB and up to a hundred files in a batch, with no account and no daily count.

For the audiences this page is written for that is mostly a convenience argument: source material for a delivery chain tends to be large, unreleased, or under a contract that says where it may be stored. Not moving it is simpler than negotiating about where it went.