Hush main

Hush

A dictation mic you hold up to your mouth, built so the people around you can't hear what you're saying. It's an aluminium cup with a sealed acoustic chamber inside. It connects as a normal Bluetooth mic for Wispr Flow or any dictation app, and holds push-to-talk while it's against your face. There are no buttons or ports; it charges on a Qi puck.

Dictation is faster than typing, but not something you can do in a library or an open office. Hush is my attempt at fixing that with hardware.

Status: design stage, nothing built yet. The exterior CAD is done, the charging puck PCB passes ERC and DRC and is ready to order, and the main board is about 95% routed. Next is printing the acoustic test rig to see if the privacy target holds up.

Specs

Privacy

Once sound is out in the room, software can't take it back, so the privacy has to be physical. A silicone cushion seals around the mouth (even a 1% leak would cap privacy at 20 dB). Behind a fabric dome there's absorber material, plus Helmholtz resonators tuned to 300 to 3,000 Hz, where foam is weakest and most of what makes speech intelligible sits. Breath escapes through a long lined labyrinth, and the aluminium wall reflects the rest back inside. The mic sits about 2 cm from your lips, so it still gets a clean signal.

PCBs

The KiCad schematics and layouts are generated from Python (design.py in each board folder), so every connection can be reviewed as code. One script runs the schematic, ERC, placement, the Freerouting autorouter and DRC, then writes Gerbers, the BOM and renders for JLCPCB.

• pcb/hush_puck: USB-C to a 5 V Qi transmitter. 20 x 20 mm, 2 layers. Done.
• pcb/hush_main: ESP32, Qi charging, fuel gauge, accelerometer and buck-boost. Ø64 mm, 4 layers, 57 parts. About 95% routed.
• pcb/hush_ring: light ring, face sensing and the two mics. Schematic only; the layout waits on the acoustic insert.

CAD

Parametric CadQuery scripts in cad/source build the cup and a separate test rig with no electronics in it; renders are done in Blender. They output STEP files for CNC, STL for printing and an 11-part print kit for the acoustic test (docs/acoustic-test.md).

Why there's a v1 folder

The first version ran speech-to-text (Moonshine) on an i.MX 8M Plus inside the cup and typed the result as a Bluetooth keyboard. When I worked out the power and heat (docs/architecture.md), it would sit at or above 43 °C against your face, last under two hours and weigh over 200 g. So v2 is just a mic, and the computer does the transcription. That also freed about 84 cm³ for a bigger acoustic chamber. The v1 board and CAD are in pcb/archive and cad/source/archive_v1, and firmware/bench is the speech model benchmark I wrote for it.

Docs

• docs/decisions.md: every design decision and change, in order
• docs/architecture.md: power, thermal and mass calculations
• docs/research/: supplier, antenna and sourcing notes
• ASTRA.md: the original design brief

Next

1. Print the acoustic rig and measure how much quieter it makes speech, and how hard it is to breathe through
2. Get Bluetooth mic plus push-to-talk working on an ESP32 dev board with Wispr Flow
3. Firmware, finish the main board and order boards