XDA describes a personal setup using an old Android device for voice input alongside Home Assistant and local models. Whisper converts speech to text, a local language model processes requests and Piper synthesizes speech. The author extends the same synthesis path through tts.speak and a queue to read persistent notifications aloud. Piper is a speech synthesizer, not a recognizer. The configuration can avoid sending voice content to the cloud, but it is a personal case study rather than a benchmark guaranteeing response time or reliability on all hardware.