v0.10.0: Sortformer

Sortformer: Real-Time Speaker Diarization

CoreML version of Nividia's Sortformer

• Real-time streaming - Speaker labels as audio comes in
• Noisy environment support - Works where traditional pipelines fail
• Overlapping speech - Scores all 4 speakers independently per frame, multiple can be active simultaneously
• Single neural model - No complex pipeline, just one model

---

Credit to @SGD2718 for the Sortformer implementation & model conversion.