Meetings
Transcription and speaker labels
Sloth transcribes meetings with speech models that run on your Mac, so your audio is never uploaded for transcription. Your voice is labelled “You”; the other side is split into “Speaker 1”, “Speaker 2” and so on.
On-device models
Meeting transcription only offers models you've downloaded to your Mac. There is no cloud option for meetings.
During setup Sloth downloads a recommended model, Parakeet Unified (English, about 565 MB), in the background and uses it for dictation and meetings. Settings › Models lists the others, including:
| Model | Size | Good for |
|---|---|---|
| Parakeet Unified | ~565 MB | English |
| Parakeet v3 | ~450 MB | 25 languages |
| Whisper (Tiny, Small, Medium, Large Turbo) | ~153 MB to ~1.5 GB | English-only or multilingual variants |
| Nemotron 3.5 Multilingual | ~665 MB | Live text in more than 100 locales |
| Cohere Transcribe | ~3.8 GB | Difficult accents and audio, 14 languages |
| Apple Speech | Managed by macOS | macOS 26; Apple's on-device model |
Pick the model for meetings in Settings › Meetings › Transcription › Final transcript. Models download the first time you choose them, from Hugging Face or Muesli's model mirror. After that they run offline.
Speaker labels
Sloth knows which audio came from your microphone and which came from the call, so labelling works without any voice training:
- Everything from your microphone is labelled You.
- After the meeting, Sloth separates the voices in the call audio and numbers them in the order they first speak: Speaker 1, Speaker 2, and so on.
- If Sloth can't separate the voices, the call side is labelled Others.
Each transcript line has a timestamp and a label, for example [00:04:12] Speaker 2: Let's move the launch.
Live transcript
On macOS 26 the transcript fills in as people talk, using Apple Speech, the on-device engine built into macOS. There is nothing to download and the audio stays on your Mac. Apple Speech writes both the live and the final transcript.
To use a different model, download it in Settings › Models and pick it in Settings › Meetings › Transcription › Live transcript model. Parakeet Live Captions (~430 MB) adds a live preview and leaves the final transcript to your Final transcript model, which is usually more accurate. Nemotron 3.5 writes both, in more than 100 locales. Set it to Off to fill in the transcript as each stretch of speech finishes. On older macOS versions, live text needs one of the downloaded models.
Hover over the floating indicator during a meeting to read the recent transcript.
Where the audio goes
- While recording, audio is written to temporary files on your Mac. Leftover temporary audio is cleaned up the next time Sloth starts.
- Sloth keeps a recording only if you ask it to. Settings › Meetings › Recording › Save meeting recording offers Never (the default), Ask every time and Always.
- Saved recordings go to
~/Library/Application Support/Sloth/meeting-recordings/, as M4A (smaller) or WAV (lossless). Choose in Recording format. - The transcript is stored in Sloth's database on your Mac. With the Sloth folder on (the default), it's also written to the
transcripts/folder.