Transcribe M4A

M4A is the most common format uploaded here — 29% of all files, more than MP3 and MP4 combined — and the numbers say what those files are. The median M4A runs 25 minutes 56 seconds at 85 kbps: that is a voice memo or a recorded meeting, not an album track.

Getting the recording out of Voice Memos

1. On an iPhone or iPad: open the recording → the three dots → Share → Save to Files, or AirDrop it to a Mac. Either way you get the original .m4a.
2. On a Mac: Voice Memos stores recordings under ~/Library/Application Support/com.apple.voicememos/Recordings, but dragging the item out of the app window is easier and does the same thing.
3. From a messaging app: forward to yourself and save. Note that some apps re-encode on send — if you have the original, use it.

Drop it into the box at the top of this page. The first five minutes need no account.

These are medians measured on our production cluster, expressed as time per hour of audio.

Against the M4A files actually uploaded here: the median one, just under 26 minutes, comes back in about 38 seconds. Nine in ten are under 65 minutes — around a minute and a half. Queue time at peak hours sits on top of these and has nothing to do with the model.

Nothing, at the container level. All three are MPEG-4 Part 14, the same box structure with the same atoms. The extension is a hint about what is inside, not a different format:

• .m4a — audio only, almost always AAC. What Voice Memos and iTunes write.
• .mp4 — the general case, usually with a video track.
• .m4b — audio plus chapter markers, the audiobook convention. Uploads here like any other.

So renaming .m4a to .mp4 generally works and is never necessary — all three upload as they are. One extension is genuinely different: .m4p is FairPlay-protected AAC from the iTunes Store, and it cannot be decoded by anything except Apple's own software. We refuse those at the door rather than accept an upload that is certain to fail.

M4A also usually holds ALAC when someone exports losslessly. That works too, and as with FLAC it buys no accuracy — the model resamples to 16 kHz mono regardless.

For M4A it is duration, by a wide margin. At the 85 kbps median we see, 5 GB would hold well over a hundred hours of audio — the 8-hour limit arrives long before the size one, and the largest M4A processed here so far was 845 MB.

• Split a long recording by session. A day of interviews is several files, not one. Cut between them rather than mid-sentence; no transcriber recovers a word split across two uploads.
• There is no need to compress further. At voice-memo bitrates the upload is already small, and re-encoding costs a generation of quality for nothing.
• Keep the original. A transcript can be re-run; a deleted recording cannot.