speaker-diarization-3.1 is an open-weight audio or speech model from pyannote.
speaker-diarization-3.1 is an open-source pipeline by pyannote for automatic speaker diarization, released under the MIT license.
It processes mono audio sampled at 16kHz, automatically downmixing multi-channel files and resampling as needed, and outputs speaker segments as an Annotation instance.
The model runs in pure PyTorch without onnxruntime, requires pyannote.audio version 3.1 or higher, and supports GPU processing.
Summary of the pyannote/speaker-diarization-3.1 model card, 2026-10-01