Repository navigation
First release of FASTDIAR: streaming speaker diarization with a fixed 960 ms delay.
Includes inference, benchmarking, and training code, plus pre-trained checkpoints of the streaming encoder:
| Model | Checkpoint | Parameters |
|---|---|---|
small |
fastdiar-small-90be5da5.pt |
2.4 M |
medium |
fastdiar-medium-32df7509.pt |
4.6 M |
large (default) |
fastdiar-large-f6d71444.pt |
10.4 M |
The checkpoints are downloaded automatically on first use; see the README for usage.