Transcription tier#
One choice for every desktop in your organization. Members see on their desktop which speech model it runs, as you set it, and can’t change it there.
| Feature | Description | Default |
|---|---|---|
| Transcription tier | How closely desktops transcribe: Fastest, Balanced, Enhanced or Best. A closer tier catches more, especially in real meetings, but works the computer harder and needs a bigger speech model, which each desktop downloads the first time it uses that tier. | Balanced |
The tiers#
- Fastest works the computer least. It’s fine for clear speech close to the microphone, and misses more when the audio is noisy. About a 75 MB download.
- Balanced is quick, and solid for everyday speech and clear meetings. About a 145 MB download.
- Enhanced is noticeably better in real meetings, with people talking over each other, accents and voices far from the microphone, for more of the computer’s effort. About a 465 MB download.
- Best is the most accurate, and the heaviest: worth it when the transcript really matters. About a 1.5 GB download.
When a desktop picks up a new tier, it stops listening while it switches. The first time it uses a tier, it downloads that tier’s speech model from Hugging Face, once, and keeps it on the computer, so that first switch takes longer.
Related#
- Meetings: what a meeting may record besides its transcript.
- Desktop awareness: what desktops record besides speech.
- Models & provider: where the text goes once it’s transcribed.