Catcher Diarization — Streaming Sortformer v2.1, MLX INT8
MLX affine INT8 (group 128) conversion of NVIDIA Streaming Sortformer
Diarizer 4spk v2.1 for the Catcher/Tippi
on-device speech runtime. Detects up to 4 speakers with 80 ms frame
resolution at 16 kHz. Converted with sortformer-convert; inference runs on
MLX-C/Metal with no Python, PyTorch, NeMo, or ONNX Runtime.
- Downloads last month
- 5
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for wcamon/catcher-diar-mlx-int8
Base model
nvidia/diar_streaming_sortformer_4spk-v2.1