WhisperX

WhisperX is an open-source speech recognition toolkit that extends OpenAI's Whisper with the two features production transcription needs most: word-level timestamps via forced phoneme alignment, and speaker diarization through pyannote.audio integration.

It also adds voice activity detection to cut hallucinations on long audio, batched inference for speed, and a faster-whisper backend option. The result became the default Whisper wrapper for subtitling, meeting transcription, and media pipelines that need exact timing.

WhisperX is released under BSD-3-Clause and remains one of the most starred Whisper ecosystem projects. It is a toolkit around models, not a new acoustic model itself, which is worth knowing when comparing it to Whisper variants.