Whisper JAX
JAX-based Whisper implementation optimized for TPU/GPU with 70x+ speedup.
About
Whisper JAX is a JAX and Flax implementation of OpenAI Whisper, built on the Hugging Face Transformers version and optimized for TPU and GPU, reporting over 70 times faster inference than the original PyTorch code on TPU. It supports batched inference and chunked processing for long audio and runs on CPU, GPU, or TPU standalone or as an endpoint. Released under the Apache 2.0 license.
Reviews (0)
Leave a Review
No reviews yet. Be the first to review!
Details
- Price
- Free
- Platform
- Local/Desktop
- Difficulty
- Advanced (4/5)
- License
- Apache-2.0
- Minimum VRAM
- 8 GB
- Added
- Apr 3, 2026
Related Tools
CLI tool that transcribes audio 10x faster using pipeline optimizations.
Self-supervised speech representation model by Meta for ASR.
High-performance C/C++ port of Whisper for CPU-based speech recognition.
Non-autoregressive ASR model by Alibaba achieving fast parallel transcription.
General-purpose speech recognition model by OpenAI trained on 680K hours of multilingual audio.
Multilingual ASR model by NVIDIA supporting 4 languages with translation.