Convert M4A to text

The advice to convert first is everywhere, so in August 2026 we tested it against our own live pipeline. We recorded a 27 second spoken memo (names, dates, a couple of money figures and some office jargon), encoded it as AAC in an M4A the way a phone does, then re-encoded that same M4A to MP3 at the same bitrate, which is exactly what conversion guides tell you to do. Both files went through the identical transcription request.

The M4A returned the surname correctly: Priya Raghunathan. The MP3 made from it returned Priya Ranganathan. Everything else matched across both runs, including March 14, April 2, 42,000, $65,000, SOC 2 and a second surname, Delacroix. The converted file also broke one sentence in a different place. It was not even smaller: 432 KB against the original's 449 KB, so the detour bought no disk space either.

Two things are worth taking from that. The first is that conversion cannot add detail that was never captured, it can only discard some, so the original file is always the safer upload. The second is more useful: both runs came back with 99.9 percent confidence, including the one that got the name wrong. A confidence score will not warn you about this, which is why the two-minute check below matters more than any number the engine reports.

Being straight about the limits of that test: it was one sample and one word of difference, not a benchmark. It points in the direction the theory already predicted rather than proving a rule.

On iPhone or iPad, open Voice Memos, tap the recording, tap the share icon, and either save it to Files, AirDrop it to a Mac, or email it to yourself. You can also skip the transfer completely: open this page in Safari on the phone and upload straight from Files, which turns your phone into the whole workflow rather than the first step of one.

On a Mac, QuickTime audio recordings and Voice Memos both land as .m4a and drag straight from Finder into the upload box. Worth knowing: macOS dictation cannot help here, because it types what it hears through the microphone live and has no way to open a file you already recorded. The same is true of Windows voice typing. Uploading the file is not the workaround, it is the actual answer.

Zoom is the other common source. Its local recordings produce an audio-only track saved as M4A alongside the video, and that audio file is the better upload: same speech, much smaller file, much faster transfer.

Speech recognition is strongest on ordinary conversational words and weakest on proper nouns, which is precisely what our test showed. Names are where corrections cluster, so read those first. If a name matters and appears throughout the recording, fixing it once in the synced editor and then searching the transcript for the other spellings takes less time than reading the whole thing twice.

Better still, prevent it. Type recurring names, company names and specialist terms into the vocabulary box before uploading, a field available on every plan including free, so the engine expects those words instead of guessing at them.

Numbers deserve the second glance. Spoken figures are converted to digits automatically, which is what you want in a memo you will act on: our test turned March fourteenth into March 14 and sixty five thousand dollars into $65,000 without being asked. That is helpful right up until a currency amount or a percentage is the one thing that has to be exact, so check those against the audio before the transcript becomes a decision.