diart

Why we included this project

Diart answers a question most speech tools skip: who is speaking, and when, while the conversation is still going. Rather than processing recorded audio in a batch afterwards, it runs a speaker segmentation model and an embedding model through an incremental clustering pipeline that gets more accurate as the stream continues. So it fits live meetings, call centers, assistive hearing, and transcription products that need answers mid-conversation. Beyond the ready-made SpeakerDiarization pipeline, you can assemble custom streaming pipelines, tune hyperparameters, benchmark results, and serve diarization over WebSockets. That combination makes it useful both as a drop-in component and as a starting point for researchers and engineers working on streaming audio, with documentation that covers installation and the main workflows.