Whisper JAX

by Sanchit Gandhi (HuggingFace)

70x faster Whisper on TPUs via JAX + Flax + batching.

Looking at Whisper JAX? Try this first.

Drop your audio. Transcript in seconds. First transcript free, then $4 = 300 min

TL;DR

Best for bulk transcription on TPU pods or A100 clusters where JAX shines. Pricing: free.

Category
Open source
License
Apache-2.0
Stars
★ 4.7k
Last push
2024-04-03
Pricing
free
Platforms
Linux, TPU, GPU

What it is

A drop-in JAX re-implementation of OpenAI Whisper that batches across the time dimension and reports up-to-70x speedups on TPUv4. MIT-licensed; the canonical choice when you have JAX infrastructure already.

Best for: Bulk transcription on TPU pods or A100 clusters where JAX shines.
Watch out for: JAX/TPU setup is non-trivial; CPU/MPS inference is slower than faster-whisper.

Install / use

pip install git+https://github.com/sanchit-gandhi/whisper-jax.git

Features

Speaker diarizationNo
Word-level timestampsYes
Streaming / real-timeNo
Languages supported99
HIPAA eligibleNo

Links

GitHub repo ↗

Whisper JAX vs Whipscribe

FeatureWhisper JAXWhipscribe
CategoryOpen sourceTranscription APIs
Pricingfree$4–$24 one-time packs (credits never expire) · $2 single unlock · 60 min free on signup
Speaker diarizationNoYes
Word timestampsYesYes
StreamingNoNo
Languages9999
PlatformsLinux, TPU, GPUWeb, API, MCP

Alternatives to Whisper JAX

OpenAI Whisper
OpenAI

The reference open-source multilingual ASR model from OpenAI.

OSS · MIT ★ 107.0k
whisper.cpp
Georgi Gerganov

C/C++ port of Whisper — runs on anything, from a Raspberry Pi to Apple Silicon.

OSS · MIT ★ 52.8k
faster-whisper
SYSTRAN

4× faster than reference Whisper using CTranslate2 — production sweet spot.

OSS · MIT ★ 24.8k

Whipscribe is a managed faster-whisper + whisperX service. If you want transcripts without running infrastructure, paste a URL or drop a file in the form below — you'll have a transcript in seconds.