whisper-diarization

Why we included this project

A transcript without speaker labels is only half useful when several people are in the recording. This pipeline connects OpenAI Whisper's transcription to a voice-activity detector and a speaker-embedding model, so each sentence in the output is tagged with the voice that said it. It strips out the vocals first to give the speaker embeddings cleaner audio, then returns a labeled, timestamped transcript instead of an undifferentiated block of text. Teams building meeting-note tools or searching through call recordings can treat it as a ready-made building block, and the included Colab notebook makes it easy to try against your own files before going further.