Convert OGG to AIFF

It is worth saying at the top, because the whole vocabulary of archival practice pushes the other way. Vorbis and Opus discard parts of a recording permanently, according to a model of what a listener will not miss. AIFF then stores every sample it is handed, uncompressed, exactly as handed over. The result is a large file containing precisely what the Ogg contained.

That satisfies a specification written around file formats. It does not protect the recording, improve it, or recover anything the original encoder removed. Where the specification is a proxy for "give us something we can still open in forty years", it is doing useful work; where it is read as "give us the best copy", the best copy is whatever the recorder originally wrote, and converting cannot manufacture one.

An AIFF is a sequence of chunks and the file produced here holds exactly two. COMM states the channel count, the frame count, the bit depth and the sample rate. SSND holds the samples. There is no name chunk, no annotation chunk, no author chunk and no artwork.

Ogg, by contrast, carries an open list of Vorbis comments, and in oral history and language documentation those fields routinely hold the speaker, the interviewer, the place, the date, the language, the project code and the consent reference. All of it is dropped. Export the comments to a sidecar — a CSV or a per-file text document — before converting, and deposit the two together, because the AIFF cannot be asked afterwards what it contains.

If the repository accepts FLAC, deposit FLAC. It is lossless, roughly half the size of the same audio as AIFF, and it carries a proper tag block so the descriptive metadata stays inside the file rather than beside it. It also holds an MD5 of its own decoded audio, so a decoder can prove years later that nothing has rotted, which no AIFF can do.

Many archives have added it to their accepted lists for exactly those reasons. Where the specification still names AIFF or WAV outright, comply — an argument in a deposit form is not a good use of anyone's time, and a rejected submission is worse than an inefficient one. But it is worth one email to the collection manager before converting several hundred hours of interviews.

Ogg is a container and the registry lists Vorbis, Opus and FLAC for it. Recordings from older open-source tooling are Vorbis; anything from a recent recorder or converter may be Opus; an Ogg holding FLAC is already lossless. Size distinguishes them — one hour is roughly 57 MB for Vorbis at 128 kbps and several hundred for FLAC.

If the container holds FLAC, converting to AIFF is a lossless step in audio terms and still discards the tags, which makes it a poor trade unless the specification forces it. If it holds Vorbis or Opus, the deposit should record somewhere that the uncompressed file was derived from a lossy original, because a future researcher looking at a 600 MB AIFF will otherwise assume it is a master.

The PCM written here is signed 16-bit. From a lossless master that would be worth examining, since a 24-bit field recording carries depth that 16 bits cannot hold. From a Vorbis or Opus source 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.

Where a specification asks for 24-bit uncompressed audio, this converter cannot produce it, and the correct response is to go back to the original recordings rather than to pad a lossy file. If the field recorder wrote 24-bit WAV before anything was compressed to Ogg, those files are the deposit the specification was written for, and they are worth looking for before converting anything.

Nothing here resamples. Whatever rate the decoder reports is the rate the AIFF gets, so a 44,100 Hz Vorbis recording produces a 44,100 Hz AIFF and a 48,000 Hz one produces 48,000. That is the correct behaviour for a deposit, since resampling is a decision with consequences and should be made deliberately.

Opus is the exception and it is not optional. The codec runs on a fixed 48,000 Hz clock and a decoder reports that rate regardless of what was recorded, so an Opus recording made at 44.1 or 32 kHz arrives as a 48 kHz AIFF. A catalogue that records the technical rate from the deposited file will therefore disagree with the field notes, and it is worth writing the original rate into the sidecar.

Uncompressed 16-bit stereo at 44,100 Hz costs 176,400 bytes a second — about 10 MB a minute — whether the minute is speech or silence, because every sample is written out as a plain number. Mono halves it. A Vorbis recording at 128 kbps costs about 0.96 MB a minute, so the multiplier is roughly ten.

A one-hour interview goes from around 57 MB to about 600 MB, and a hundred hours of fieldwork from under 6 GB to nearly 60. That is the number to take to whoever is paying for the storage, and it is the strongest practical argument for asking whether FLAC would be accepted instead — the same lossless-in-form deposit at half the size.

A great deal of documentation audio is genuinely single-source: one lapel microphone, one handheld recorder on a table. Converting does not promote it to stereo, and it should not — two identical channels double the file for no information. The COMM chunk records the count faithfully, so a mono recording arrives as a mono AIFF.

Where two microphones were used and the channels carry different speakers, that separation is often the most useful technical property the recording has, because it makes a transcript attributable. Nothing here folds it down. What the AIFF cannot record is what the channels mean: COMM states how many there are and not which microphone was where, so that belongs in the sidecar with everything else.

This is not an incidental property for this reader. Interviews are conducted under consent forms that name the project and frequently prohibit disclosure to third parties, and passing a file through a conversion service that holds it for an hour is a disclosure whether or not anyone reads it.

Nothing here goes anywhere. Writing AIFF costs no dependency — the samples are muxed as a WAV and rewrapped by arithmetic — and the Ogg decoder is already on your machine, so no package is downloaded and no audio is transmitted. Files are capped at 100 MB each, which at 128 kbps is about an hour and three quarters of Vorbis, and a hundred can be dropped at once and returned as a ZIP.