Transcription library/cli/server base on OpenAI Whisper model written using Burn framework 🔥.
-
Simple Whisper lib: Implements the Whisper model via the Burn WGPU backend. Based over sudomonikers/whisper-burn, provides abstraction over different models and languages. Weights are automatically downloaded from Hugging Face repo. Supported codec: flac, vorbis, wav, mp3
-
Simple Whisper cli: CLI application useful to transcribe audio file. For more information see the README.md.
-
Simple Whisper server: Websocket server that transcribe uploaded files.
- Show how malleable RUST is, scaling from server to GPU code.
- 0 build prerequisites.
- Support a high variety of platforms.
- Fast enough on every platform.
- It is NOT intended to be the fastest/accurate Whisper implementation.
- NOT production ready
The project was inspired by:
- Previous whisper Burn implementations: sudomonikers/whisper-burn and Gadersd/whisper-burn.
- Candle implementation: rwhisper.