A talking desk companion can be planned around a Raspberry Pi-class computer, microphone array, speaker, camera and two or three servos for an expressive head. This is a proposed DIY architecture, not a RoboHub-tested build. Begin with a stable base and limited head travel; get audio working before adding motion or optional camera features.
Design the voice pipeline as wake word, speech-to-text, language model and text-to-speech. Choose local or cloud speech/language models after measuring response time on your hardware. Local processing gives more control over audio; cloud processing introduces network dependence and potential usage costs. Neither route guarantees instant replies. Set request limits, show when the mic is active, and avoid recording by default.
For a kit reference, Pollen Robotics describes Reachy Mini Wireless with an onboard Raspberry Pi CM4, Wi-Fi and USB, camera, four microphones, speaker and accelerometer. Lite uses a Mac, PC or Linux computer over USB with the same motion capabilities. Both are DIY kits with about two hours of assembly. The maker describes built-in realtime-model conversation and more than 60 community apps; those are Reachy Mini features, not features of this proposed DIY head.