WhisperX
lastmod 2026-07-02created 2023-01-026 edits
A Whisper-based ASR with improved timestamp accuracy
Categories
- meeting_minutes_pipeline.py - Code for organizing meeting minutes
Features
- OpenAI Whisper produces very accurate transcriptions, but its timestamps are at the utterance level (Utterance) rather than the word level, so they can be off by a few seconds
- Improves the timestamps of the Whisper model using a phoneme-based ASR (Phoneme-based) such as Wav2vec2.0 and forced alignment (Forced Alignment)
- Default languages provided: {en, fr, de, es, it, ja, zh, nl}. Additional languages need to be found on the Huggingface Model Hub and tested
See also
- Deep Learning
- Whisper
- Nightingale - Open-source karaoke app that can use any song on your computer
- whisper-diarization