A developer has created a fully local home automation voice assistant, named "Fulloch," using Qwen3 models for automatic speech recognition (ASR), large language model (LLM), and text-to-speech (TTS).

Implementation Details

The system runs on an RTX 5060 Ti graphics card equipped with 16GB of VRAM. The video demonstration showcases latency and response times using Qwen3 (ASR and TTS 1.7B, Qwen3 4B Instruct 2507) with a voice clone. The project includes tools to control devices such as Philips Hue, AirTouch climate control systems, and online weather information retrieval (specific to Australia).

Alternative Models

Smaller models were also tested for intent generation, but response quality dropped dramatically with LLM models smaller than 4 billion parameters. Kokoro (TTS) and Moonshine (ASR) are included as options for systems with limited resources.