I've been shocked by the amount of money people are paying for hardware transcription devices and subscriptions for the software.

I built EarWing, which you can 3D print and assemble yourself. It uses a ESP32 dev board and 4 microphones for noise cancellation and has a browser offload for transcription and summary generation (using LLMs running in the browser).

The software if fairly rough at the moment but works quite well in Chrome. It uses Parakeet for transcription and LFM 2.6B for key point, open question and action point extraction.

It's all open source and available at https://github.com/nlothian/earwing