TOOLORA / Local reading edition
WhisperX
Combine batched ASR, phoneme forced alignment and speaker diarization to create transcripts with word timestamps and anonymous speaker labels.
Overview
WhisperX chains faster-whisper transcription, VAD segmentation, language-specific phoneme alignment and pyannote diarization for word timelines, subtitles and speaker-labeled transcripts. Recognition, alignment and diarization are separate stages. The current pyannote model requires accepting its terms and providing a Hugging Face token. Overlapping speech, out-of-dictionary words and numbers may not align accurately; speaker labels do not identify real names.
Official sources and references
- Official websitehttps://github.com/m-bain/whisperX
Descriptions summarize public information; consult official websites for features, pricing and regional availability.