Voice-Decibel Diary Device

I want to build a small, always-on unit that sits in my living room, measures the decibel level of anyone speaking, recognises who is talking, and stores a running “voice diary” for the entire day. The end goal is a private log that shows, for every utterance, the person detected, how loud they were, the exact time, and how long they spoke.

Key behaviour
• The device must learn voices passively from normal conversations rather than asking family members to record enrolment samples. Over time it should refine its speaker models automatically.
• Accurate decibel measurement is essential; readings should be reliable across different positions in an average-sized room.
• All data stays local. The diary can be exported (CSV or SQLite is fine) but nothing should be sent to the cloud.

Typical stack I envision: a low-power board such as Raspberry Pi or ESP32, a calibrated microphone array, Python/C firmware, and a lightweight speaker-recognition model (e.g., Vosk, PyAnnote, or similar). Feel free to suggest alternative hardware or libraries if they achieve solid voice-ID accuracy without heavy resources.

Deliverables
1. Hardware part list and wiring diagram.
2. Ready-to-flash firmware / software with setup instructions.
3. Voice-recognition model that self-adapts, plus decibel-logging logic.
4. Local database schema and a simple dashboard or script that lets me filter by person, volume range, date, and total speaking time.
5. Short test report showing detection accuracy for at least three speakers across quiet and loud scenarios.

I’ll handle enclosure design and long-term power, but the electronics, code, and data handling are yours. Success is a plug-and-play prototype I can power on, let run for a day, and then open the log to see who spoke, how loudly, and for how long.