Transcribed on your Mac
Riff turns speech into text on your Mac, for dictation and for meetings, with the engine you choose for each.
Two engines
Apple Speech is the default and is part of macOS. Parakeet Ultra is an optional download in Settings under Models, with its size, a Download button, progress you can cancel, and Delete.
| Engine | What it is | Text while you talk | Languages | Download |
|---|---|---|---|---|
| Apple Speech | The speech recognition built into macOS | Yes | More than 45 | The speech model for your language, which macOS downloads |
| Parakeet Ultra | Moondream’s Parakeet Ultra, built on NVIDIA’s Parakeet | Yes | 25 European languages | About 1.2 GB |
You choose one engine for dictation and one for meetings. Both write the transcript as you talk.
Apple Speech
Apple Speech uses the speech recognition in macOS. Settings under Models shows the state of the speech model for your language. When your dictionary has words and your language supports it, Riff passes them to the recognizer as hints.
Parakeet Ultra
Riff downloads Parakeet Ultra once, and Iterator’s other Mac apps use the same copy. It hears which of its 25 languages you speak, so you don’t set one.
Speaker labels
Speaker labels is a download of about 250 MB in Settings under Models, and it works with either engine. When a meeting ends, Riff tells apart the voices on the other side of the call, and the people around your Mac when more than one of you talks into its microphone. Riff doesn’t guess which voice is yours. Each one reads Speaker 1, Speaker 2, and so on until you name it in the transcript, where Riff suggests names from the calendar event. An audio or video file you import gets its voices told apart the same way.
What happens next
The text goes through your replacements and snippets, then cleanup. For dictation, Polished cleanup and edit by voice use a writing model. For meetings, the transcript and your notes go to a writing model after the call. Writing on your Mac explains where that runs.