pyannote.audio

pyannote.audio is a Python toolkit for developers and researchers who need to separate recordings by speaker. Speaker diarization identifies who spoke when, so applications can associate parts of an audio recording with different speakers.

• Speech segmentation: Detect speech activity, speaker changes and overlapping voices as part of separating audio into speaker turns.
• Speaker embeddings: Generate voice representations for speaker recognition and verification tasks.
• Model adaptation: Use pretrained PyTorch models and pipelines, or fine-tune them on your own audio, with multi-GPU training through PyTorch Lightning.

The toolkit is open source under the MIT license. It supports loading models from local files or directories, and pretrained models and pipelines are available through Hugging Face. Its Python interface also supports paid precision-2 diarization through pyannoteAI, which requires a pyannoteAI API key.

Optional telemetry sends anonymous usage metrics, including model or pipeline origin, audio duration and speaker-count settings. You can disable telemetry.