On-Device Offline Voice Assistant with Whisper Tiny & Quantized LLM
100% offline, privacy-first smart home assistant running quantized Whisper ASR and TinyLlama on a local single-board computer.
Project Overview
Requires zero cloud connectivity, safeguarding household privacy. Uses an optimized onnxruntime Whisper Tiny model to transcribe voice commands into text in under 200ms, parses home automation intents with a 4-bit quantized local LLM, and triggers local relays/lights.
Requires zero cloud connectivity, safeguarding household privacy. Uses an optimized onnxruntime Whisper Tiny model to transcribe voice commands into text in under 200ms, parses home automation intents with a 4-bit quantized local LLM, and triggers local relays/lights.
100% offline, privacy-first smart home assistant running quantized Whisper ASR and TinyLlama on a local single-board computer.
Core Project Objectives
Capture continuous analog/digital sensor readings with robust noise filtering and hardware calibration.