AdPet

ESP32-S3 SuperMini desktop pet using an SSD1306 OLED, INMP441 microphone, and MAX98357A speaker amp.

Current Mode

The project now runs the gateway-based voice-chat flow:

voice trigger
  -> record WAV from INMP441
  -> POST WAV to AdPet Gateway
  -> Gateway handles STT / LLM / TTS / memory
  -> play WAV through MAX98357A
  -> update OLED emotions

The ESP32 no longer talks directly to STT/LLM/TTS providers. See docs/gateway.md.

Wiring

All GPIO assignments are centralized in src/board/BoardPins.h. When the enclosure layout changes, edit that file only and rewire the matching signals. Duplicate active GPIO assignments cause a compile-time error.

OLED:

OLED	ESP32-S3 SuperMini
VCC	3V3
GND	GND
SDA	GPIO8
SCL	GPIO9

INMP441:

INMP441	ESP32-S3 SuperMini
VDD	3V3
GND	GND
SCK/BCLK	GPIO4
WS/LRCLK	GPIO5
SD/DOUT	GPIO6
L/R	GND

MAX98357A:

MAX98357A	ESP32-S3 SuperMini
VIN	5V
GND	GND
BCLK	GPIO10
LRC	GPIO11
DIN	GPIO12
SPK+	Speaker +
SPK-	Speaker -

Setup

1. In Arduino IDE select ESP32S3 Dev Module.
2. Set the flash size to match the module marking. If the module includes PSRAM, select its matching QSPI PSRAM or OPI PSRAM; otherwise select Disabled.
3. Upload AdPet.ino with Arduino IDE.
4. Open Serial Monitor at 115200 and check the [Memory] line for psram=yes/no.
5. Connect your phone/computer to Wi-Fi AP AdPet-Setup, password adpet1234.
6. Open the setup page printed in Serial Monitor, usually http://192.168.4.1.
7. Fill Wi-Fi and Gateway settings.
8. Save config and reset the board.

Default gateway settings:

Gateway Base URL: http://192.168.1.100:8787
Device ID: adpet-001

Runtime

When the microphone level crosses the configured threshold:

1. OLED changes to the configured listening expression.
2. ESP32 includes about 300 ms of pre-roll audio so the first word is not cut off.
3. Recording ends after about 700 ms of silence or at the configured maximum duration.
4. Gateway receives WAV and handles STT, LLM, TTS, and memory.
5. Gateway returns WAV audio and an emotion name.
6. The speaker plays the reply, then microphone triggering resumes after a short guard period.

Serial Monitor prints each step, including HTTP status codes.

Without PSRAM, recording is capped at 3000 ms to preserve memory for Wi-Fi and HTTP. With PSRAM enabled, recordings, HTTP request bodies, and TTS WAV data prefer PSRAM.

Project Structure

AdPet.ino
src/app              app orchestration
src/board            all hardware GPIO assignments
src/core             config, emotions, pet state
src/drivers/audio    INMP441/MAX98357A audio
src/drivers/display  SSD1306 faces
src/services/network Wi-Fi and web config
src/services/llm     AdPet Gateway client