whisperX

Why we included this project

Whisper gives you a transcript, but not the word-level timing that subtitles and search tools need. WhisperX fixes that by adding forced phoneme alignment and voice-activity detection on top of OpenAI's ASR model, so each word gets a timestamp that actually lines up with the audio, and it batches inference to run fast. That per-word precision matters for subtitle generation, meeting notes, and media search, where you want to jump straight to the moment a phrase was spoken. It also layers in speaker diarization via pyannote, so multi-speaker recordings come out labeled by who said what. A CLI and Python API keep it easy to slot into existing transcription pipelines, and the alignment models cover a useful range of languages out of the box.