WhisperX 🧩 Workflow Open source

What it is

WhisperX by Max Bain (BSD-2-Clause) builds on OpenAI's Whisper to give fast batched transcription, accurate word-level timestamps through forced alignment, and speaker diarization via pyannote. It is the engine behind many subtitle, clipping and dubbing tools, and a solid base for your own transcription pipeline for podcasts and videos. Where plain Whisper gives only rough segment times, WhisperX aligns each word to the audio, which is what caption burn-in, clip cutting and dubbing tools need to stay in sync. Speaker labels make interview and podcast transcripts far easier to edit.

Key features

• Batched inference with faster-whisper for high speed
• Word-level timestamps via phoneme alignment
• Speaker diarization with pyannote models
• Voice activity detection to reduce hallucinations
• Command-line tool and Python API
• Outputs SRT, VTT, JSON and text

How to install

Run pip install whisperx (or clone and uv sync --all-extras --dev for development). Diarization needs a Hugging Face token and acceptance of the pyannote model terms. A CUDA GPU is recommended.

Pricing and rights

Free under BSD-2-Clause. Whisper models are MIT-licensed; the pyannote diarization models require accepting their terms on Hugging Face. Runs locally, so no per-minute fees.

Who it is for

Podcasters, video editors and developers who need accurate, timestamped transcripts and subtitles, or who are building their own clipping and dubbing tools.

Verdict

A dependable workhorse for local transcription with precise timing. Development has slowed (last push July 2026) and dependency conflicts with PyTorch and pyannote versions are common.