whisperX
Fast automatic speech recognition featuring forced word alignment and speaker diarization.
Batched Whisper transcription paired with forced phoneme alignment and speaker diarization.
WhisperX provides fast automatic speech recognition by pairing a batched faster-whisper backend with phoneme-based forced alignment. It includes speaker diarization capabilities through Pyannote models. It suits voice producers, audio editors, and developers who need accurate word-level subtitle timing and speaker identification.
What it is
WhisperX speeds up OpenAI Whisper inference up to 70x real-time using VAD-based batching via the faster-whisper backend. It performs forced alignment using phoneme models like wav2vec2 to generate accurate word-level timestamps. It also integrates speaker diarization pipelines to assign speaker labels to transcribed audio segments.
Who it's for
This tool suits audio engineers, podcast producers, and software developers who require precise transcript alignments or speaker identification in their voice pipelines. It does not suit non-technical users seeking a standalone desktop interface without command-line or Python execution.
Key features
Batched Inference Backend
Uses VAD-based segmentation and faster-whisper to achieve up to 70x real-time transcription speed on GPUs.
Phoneme Forced Alignment
Aligns transcripts with language-specific wav2vec2 models to produce accurate word and character timestamps.
Speaker Diarization Integration
Integrates Pyannote speaker diarization pipelines to identify and label separate speakers in audio.
Reduced Hallucination Settings
Turns off previous text conditioning by default to prevent transcription loops and repeating text output.
Up and running in 3 steps
Our honest verdict
What's great
- Achieves up to 70x real-time transcription speed using VAD batching on GPUs.
- Provides precise word-level timestamps using forced phoneme alignment models.
- Reduces text hallucination loops by disabling previous text conditioning by default.
- Supports sentence-level segmentation and SRT outputs with word highlighting.
Where it falls short
- Requires high-end GPU VRAM to run larger ASR and alignment models smoothly.
- Demands a Hugging Face account token and manual agreement acceptance for diarization models.
- Fails to align words containing characters missing from the alignment dictionary, such as numbers or symbols.
- Struggles to accurately transcribe and separate overlapping speech.
Pricing & license
WhisperX is free and open-source under the BSD-2-Clause license. The primary cost comes from providing your own GPU hardware or compute infrastructure. Third-party cloud API hosting on Replicate is also available as a paid option.
Open remote AI-training roles
876 open remote AI-training roles, updated November 15, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Does WhisperX require a dedicated GPU?
You can run WhisperX on a CPU using int8 compute types, but achieving up to 70x real-time speeds requires an NVIDIA GPU with CUDA.
Why are some numbers or symbols missing precise timestamps?
Words containing characters not found in the alignment model dictionary cannot be phonetically aligned, leaving them without specific timestamps.
