0.2.7: Diarization Improvements
- New default backend: Sortformer is now the default diarization backend, replacing Diart
- 6x faster processing: Reduced latency from ~2s to ~0.3s on CPU
- Significantly improved speaker detection (Constraint: Currently supports up to 4 speakers maximum)
- Shared model loading: A single Sortformer model
SortformerDiarizationis now shared across users and instances to reduce memory footprint. Speaker caches, frames, etc. are handled per user inSortformerDiarizationOnline - Enhanced alignment: Improved time and token synchronization between transcription and diarization results