Speakrail - a low-latency fully-local voice assistant that runs on a single RTX 4090

Speakrail is a new open-source, fully-local, low-latency voice assistant designed to run on a single RTX 4090. It utilizes Voxtral Realtime, a microturn-finetuned Gemma 4 12B model, and Breeze TTS 2. The developer claims it rivals GPT-Live on some benchmarks. A demo shows its current capabilities, including a limitation where it can only check the weather in Moscow, despite repeated user attempts to query Singapore. The developer encourages users to try it out and provide feedback or feature requests via GitHub issues.