Tap Transcribe on a sleep-talking clip and the words appear under it, written out on your iPhone.
Snore Timeline already saves every stretch of sleep talking it hears as its own episode, so you can play it back in the morning. Transcription goes one step further. Tap Transcribe on a clip and you can read what you said, without putting on headphones and replaying a mumble five times. A speech model on your iPhone does the work, and your recordings stay on the device.
Each Sleep Talking episode holds one or more clips of speech. When you tap Transcribe, the app feeds those clips to a speech recognition model that runs on your phone and writes the words under each clip. The audio stays where it is, so you can read the text and play the clip side by side.
Nothing gets uploaded. Once the model is on your phone, transcription works in airplane mode, and no server, mine included, ever hears your recordings. Sleep talk can include names and things you'd never say awake, which is why the app keeps it all on the device.
Transcription is part of the free app. There's no subscription and no per-clip charge.
The speech model is too large to ship inside the app, so the first time you tap Transcribe, the app asks before downloading it. The sheet shows the exact size for your phone. iPhones with 6 GB of memory or more get the larger, more accurate model, 574 MB. iPhones with less memory get a smaller one, about 265 MB.
You download it once and the app reuses it every night after. The download keeps going if you leave the app, and you can follow it or cancel it in Settings > Detection Settings, under Sleep Talking, in the Sleep Talk Transcription row. When it finishes, the app installs the model and transcribes the clips you asked for.
This download is the only time transcription needs a connection.
You can also transcribe one clip at a time. Tap the chevron next to the signal count to open the clip list, then tap the transcribe button beside a clip. A short clip takes a couple of seconds on a recent iPhone. The first transcription after the download takes a little longer while the app loads the model.
The app saves each transcript with its episode, so it's there the next time you open the night.
Each clip gets one of three results:
A lot of sleep talking is mumbled, whispered or half-formed. The app marks those clips as unrecognizable rather than filling them with invented sentences, which speech models tend to do with unclear audio. The recording stays available to play back either way, and your ear is often better than any model at a 2 a.m. mumble.
The app transcribes in the language you use Snore Timeline in, which is your iPhone's language unless you've set a different one for the app in iOS Settings. The model covers about 99 languages. If yours isn't one of them, the app transcribes in English.
If you talk in your sleep in a different language from the one your phone uses, change Snore Timeline's language in iOS Settings > Snore Timeline > Language before you transcribe.
The model takes up space on your phone between uses. To free it, go to Settings > Detection Settings and tap Remove in the Sleep Talk Transcription row. Transcripts you've already made stay with their episodes. Tap Transcribe again later and the app offers the download again.
