Voice-Controlled Lights on a Raspberry Pi
summary
A complete local voice-controlled lighting system runs on a Raspberry Pi 5 (8GB) and controls bulbs via UDP; audio capture uses Membrane, the dashboard is Phoenix LiveView, and storage uses ChickenLake. Wake-word detection uses tiny ONNX models with Ortex for fast edge inference while speech-to-text uses Whisper via Nx/Bumblebee (XLA on Pi shows ~2–3s latency; ONNX tiny models are much faster). All processing can run offline, models and code are available in the linked repo so the prototype actually works at home.
AI-written summary · about
chapters (23)
- 0:00From ice to Elixir why we need a new narrative
- 2:52Introducing Nelly
- 4:34Nelly’s features and setup
- 6:19Phoenix LiveView dashboard
- 7:48Figma and Tailwind design
- 8:11Voice assistant system steps
- 9:59OpenAI API voice control
- 12:07Nx and Ortex local control
- 13:56Wake word on edge devices
- 16:02Nx for matrix operations
- 17:51Whisper model with Bumblebee
- 19:19Demo and Pi limits
- 19:40AI project lessons
- 22:40Q&A
- 23:02Raspberry Pi setup
- 24:57Custom wake words training
- 25:59Edge AI industrial uses
- 27:09Supported languages
- 28:16Long command handling
- 30:06Multi-language trade-offs
- 31:30Fuzzy logic matching
- 32:36Structured voice output
- 33:58Closing remarks
related talks
37mVoice-controlled Home Automation from Scratch Using IBM Watson, Docker, IFTTT, and Serverless
24mHow I ended up automating my curtains and shouting at my laptop We encounter automation every day and may not realise
27mHow I Ended Up Automating My Curtains And Shouting At My Laptop
36mMyCroft
58mVoice Control Made Simple – Panel Discussion
17mBuild a smart IoT device with TensorFlow Lite and Raspberry Pi
47mCreating a Connected Home
45mComparison of Voice Assistant SDKs for Embedded Linux Devices