Convert AAC to AIFF

A `.aac` file is Audio Data Transport Stream framing: frames one after another, each repeating the sample rate and channel configuration in a short header, with no container around them. That design exists so a stream can be cut anywhere and joined to anything, and the price is that nothing in the file states how long it is.

Software copes by estimating — scanning the whole file, or dividing its size by an assumed bitrate — and variable-rate material defeats the arithmetic. The result is an import that reports the wrong duration, a waveform that will not seek, or a clip that arrives on a timeline at the wrong length. The AIFF written here states its frame count outright in the COMM chunk, and for a good many readers that, rather than the compression, is what this conversion actually fixes.

AAC compression removed parts of the recording permanently, according to a model of what a listener would not miss. AIFF then does the opposite job perfectly: it stores every sample it is handed, exactly as handed over, with no compression at all. Put together, the output is a large file holding precisely the audio the AAC held.

That is worth having when something downstream requires PCM, and it is not a quality gain by any measure. Delivery specifications name uncompressed formats so that everything arriving is in a known state and nothing is re-encoded a second time in the plant. If a better source exists — the session, the master, the original bounce — that is what should be delivered, and one email is cheaper than any conversion.

An AIFF is a sequence of chunks and the one produced here holds exactly two. COMM states the channel count, the frame count, the bit depth and the sample rate. SSND holds the samples. Nothing else is written — no name, no annotation, no author, no timecode reference, no loudness measurement, no artwork.

Broadcast and library specifications frequently ask for those things in the same paragraph that asks for uncompressed audio, and a converter cannot supply them. If the delivery sheet requires an integrated loudness figure, a start timecode or an originator field, the file has to pass through an editor that can write them. Knowing that before delivery is better than a rejection notice a week later.

Nothing here resamples. Whatever rate the ADTS header declares is the rate the AIFF gets. Audio produced for broadcast, video or conferencing is normally 48,000 Hz; audio destined for a CD or originating from one is 44,100. Both convert without complaint and both stay where they were.

The mismatch matters at the two ends of the chain. An audio CD is 44,100 Hz and nothing else, so a 48 kHz deliverable will be refused or silently resampled by the authoring tool. A station running its plant at 48 kHz will resample a 44.1 file on ingest, which is a second processing step nobody asked for. Read the rate off the source before converting a batch, because a mixed set is the version of this that surfaces late.

The PCM written here is signed 16-bit. From a lossless master that would be a decision worth examining, since a 24-bit source carries depth 16 bits cannot hold. From an AAC file it is not a decision at all — a lossy codec has no additional depth to preserve, and writing 24-bit output would produce a file half again as large with no more information in it.

Sixteen bits at 44,100 or 48,000 Hz is also what most delivery specifications for this kind of material name. Where a sheet asks for 24-bit uncompressed audio, this converter cannot produce it and a desktop tool is the right instrument — but it is worth asking what the 24 bits would contain, given that the source was a compressed stream.

Uncompressed 16-bit stereo at 44,100 Hz costs 176,400 bytes a second — about 10 MB a minute — whether the minute is dense or silent, because every sample is written out as a plain number. AAC at 128 kbps costs about 0.96 MB a minute and at 256 kbps about 1.9.

So a 30-second commercial goes from roughly half a megabyte to about 5 MB, a five-minute item from 5 MB to about 50, and an hour-long programme from 57 MB to around 600. There is no quality control on this conversion because there is nothing to trade — AIFF performs no compression at all, which is precisely why specifications ask for it.

An AAC encoder needs a short run-up before it produces valid output, and a container normally records how much of the decoded result to discard so that playback starts where the sound does. A bare ADTS stream has nowhere to record that, and uncompressed PCM has nowhere either.

So a small amount of the encoder's own padding — a few dozen milliseconds — arrives as ordinary audio at the start of the AIFF. For a programme item on a timeline it is invisible. For a commercial cut to an exact length, a station ident or anything that has to hit a frame boundary, it is the first thing to check, and trimming it in an editor after converting is a ten-second job that avoids a rejected delivery.

AAC handles far more channels than AIFF does — the registry records 48 against 8 — so a wide multichannel stream has nowhere to go here, while mono, stereo and everything up to 7.1 passes through with its channels intact. Mono stays mono rather than being doubled into a fake stereo pair, which is the correct outcome for a voice track or a phone-line contribution.

What the AIFF records is the number of channels and not what they mean. The COMM chunk states a count; there is no field for a speaker assignment or a channel order. A multichannel deliverable therefore has to be checked against the specification's expected order rather than assumed, and a stereo pair that was really two unrelated mono feeds will look identical to a genuine stereo mix.

Writing AIFF costs no dependency. The decoded samples are muxed as a WAV and rewrapped by arithmetic — the header rebuilt big-endian, the sample rate written as the 80-bit extended float the format insists on, each sample's bytes reversed. Unlike the MP3, AAC and FLAC targets on this site, nothing is downloaded at all, and the AAC decoder is already on your machine.

Files are capped at 100 MB each and a hundred can be dropped at once, coming back as a ZIP. For unaired commercials, embargoed programme material or anything under a client confidentiality clause, the useful property is that this is verifiable rather than promised: open the network tab, convert a file, and watch nothing move.