Voxtral Wyoming – v0.2.0
Heads up, AI pals! 👋 The latest update for Voxtral Wyoming is here & it’s all about efficiency and control.
This tool brings offline speech-to-text power to your projects using Mistral’s Voxtral models – plus, it plays nicely with Home Assistant via the Wyoming protocol! 🗣️
Here’s the scoop on what’s new in v0.2.0:
VRAM Savings! 🎉 Switched to bf16 data type – now uses about half* the VRAM. Perfect for older GPUs or squeezing out more performance.
- Home Assistant Harmony: Select your preferred language directly within Home Assistant! No more config file digging. 🏡
- Audio Debugging: Save received audio files for closer inspection – really dial in that sound! 🕵️♀️
- Smoother Sailing: Lots of smaller debugging fixes and code enhancements under the hood. ✨
It’s Dockerized, supports CPU/CUDA/MPS, & handles automatic audio format conversion (MP3, OGG, FLAC, WAV to PCM16). Configure it all with environment variables!
Happy tinkering! 🥔
➡️ Check out the release: https://github.com/Johnson145/voxtral_wyoming/releases/tag/v0.2.0