Convert AAC to FLAC

Most people arrive at this pair because a form said so. Distributors ask for lossless masters. Podcast networks specify WAV or FLAC for the mix they will process. Mastering engineers ask for anything but MP3. Broadcasters publish delivery specs with a codec list. Stock-audio libraries and speech-dataset submissions do the same. In every one of those cases the requirement is a file format, and FLAC satisfies it.

What the requirement is trying to achieve, of course, is that the audio has not been through a lossy encoder. Yours has. Those two facts can both be true and the sensible response is not to hide one of them: convert, submit, and state in the accompanying notes that the source was AAC. Reviewers deal with that constantly and would far rather be told than find out.

A lossy encoder does not spread its damage evenly. It applies a low-pass, so the very top of the frequency range simply stops — a hard horizontal shelf at 16, 19 or 20 kHz depending on the bitrate — and that edge is unmistakable in any spectrum analyser. Audacity, iZotope RX, Spek and Sonic Visualiser all show it in the time it takes to open the file.

That is worth knowing before you send anything, because it removes the temptation. A FLAC made from AAC will pass an automated format check and will not pass an engineer with an analyser, and the professional move is to declare the provenance in advance rather than to be asked about it afterwards. Where the spec exists because the audio will be re-encoded downstream, a declared lossy source is often accepted; where it exists because the audio will be mastered, it usually is not, and knowing that early saves a rejected delivery.

When a spec allows either, FLAC wins on three counts. It is lossless in exactly the same sense — decode a FLAC and you get back the identical samples — while typically occupying half the bytes, which matters when a delivery is an hour of audio over a slow upload. It carries a checksum, so a file damaged in transfer or on a disk is detected rather than played back wrong; the registry records that property and it is the reason archives prefer it.

And it holds metadata properly. Vorbis comments carry the full set of descriptive fields and PICTURE blocks carry artwork, where WAV's RIFF INFO chunk manages seven fields in Latin-1 and no images at all. The only reason to choose WAV over FLAC is an ingest system old enough to accept nothing else, and specs that say so usually say so explicitly.

FLAC compresses by predicting each sample from the ones before it and storing only the error. That works well on a real recording, where the waveform is continuous and well behaved, and less well on audio that has been through a psychoacoustic model, which has already had its predictable structure rearranged.

The practical consequence is that decoded lossy audio compresses to roughly 60 to 70 percent of its uncompressed size rather than the 50 to 55 percent a real recording manages. A four-minute AAC at 256 kbps is about 7.7 MB and becomes a FLAC of 20 to 30 MB — three to five times larger, holding not one additional detail. Budget for that before converting an hour-long programme.

It is easy to read "lossless" as a claim about quality, and it is not one. It is a claim about the future: from this file onwards, nothing further will be discarded. Copy it, move it, open it in an editor, save it again, and the samples are identical every time.

That is genuinely valuable for a delivery file, because a delivery file gets handled. It is opened, trimmed, level-matched, spliced into a programme, exported and re-imported. Every one of those steps in a lossy format costs a little more of the recording; none of them cost anything in FLAC. So the conversion does not improve what you have, and it does stop what you have from getting worse in somebody else's hands.

The FLAC keeps the sample rate and channel layout of the source. Music from a stream is normally 44,100 Hz stereo; audio taken from video is 48,000 Hz; a broadcast or telephony capture may be 32,000 Hz or lower and mono. None of that is changed, which is correct — resampling to hit a number in a spec would add a second processing stage and no information.

If the spec demands a specific rate or depth that your source does not have, resampling has to be done deliberately, in a tool that lets you choose the algorithm, and it should be done once. It cannot be done meaningfully by a converter that does not know whether the target is a mastering chain or a phone. Convert to FLAC here, resample there, and keep the AAC as the thing you can always go back to.

AAC allows up to 48 channels and FLAC up to 8, which the registry records, and the gap is where a broadcast delivery can come unstuck. Ordinary stereo and 5.1 material passes through without incident. A layout beyond eight channels has no FLAC representation at all.

If a source configuration cannot be encoded, the conversion falls back to two channels at 48,000 Hz rather than failing outright — so an exotic multichannel capture may arrive as a stereo fold-down. Check the channel count of the first output against the source before you submit a multichannel delivery; a spec that asked for six channels and received two is a rejection that costs a week.

Where the AAC came from decides what tags exist. A file extracted from an M4A or carrying an ID3v2 block at the front arrives with a full set of fields, and all of them map onto Vorbis comments, with any embedded artwork becoming a PICTURE block.

A raw ADTS capture — a stream recording, an HLS join, camera or broadcast audio — normally carries nothing, and the FLAC comes out untagged. That is faithful rather than lossy, and for a delivery it is usually irrelevant, because the spec will want its own fields filled in anyway: ISRC, title, artist, date, the identifier the recipient uses. Add them in a tag editor after conversion, in the form the spec asks for.

The audio in a delivery is very often the thing that must not leak: an unreleased track, an episode with a publication date, an interview under embargo, a client recording covered by an agreement that names who may hold a copy. A conversion site that takes an upload creates exactly the copy that agreement was written about.

Here the AAC is decoded by your own device and the FLAC is written by an encoder running inside the tab, loaded on demand and used nowhere else. No request carries the file, there is no account and no record of what was converted. Files up to 100 MB are handled, batches come back as a ZIP, and the whole claim is checkable in the network tab of your developer tools while a conversion runs — which is the only form in which such a claim is worth anything.