An Affordable Voice Assistant That Won't Spy on You (2022) (blog.tjll.net)

A DIY voice assistant built from affordable, repurposed hardware demonstrates a practical, privacy-first alternative to Alexa or Google Assistant: the author assembled a fully local system (no cloud traffic beyond your LAN) using a Google AIY Voice Kit v1 as the microphone/speaker, a Raspberry Pi 3 as the wake-word “satellite,” and a more powerful base station (ODroid H3 or any always-on x86/laptop) running Rhasspy to do the heavy speech recognition. The result reportedly delivers usable, low-latency recognition and hands-on privacy control — a compelling option for users unwilling to route voice data through third-party cloud services. Key technical points: the architecture uses an MQTT broker (e.g., mosquitto) to bridge the Pi satellite and Rhasspy base in a multi-machine “satellite + base” topology, with Rhasspy and components run in Docker. The Pi handles wake-word detection while the base crunches audio and intents; this avoids the Pi Zero/v2 latency problems and keeps expensive models off the SBC. Integrations shown include Mopidy/MPD, Kodi, blinds (soma-ctrl), timers, Anylist, and a custom Python daemon (with an “undo” stack). The guide notes hardware sourcing challenges (AIY v1 and Pi3 often via eBay), power requirements (5V/2.5A), and points to Whisper as promising for future on-device ASR but not yet central to this local-first stack.