(Arduino + AI Voice Therapy Bot)

A compact, voice-activated AI companion built using the Arduino UNO R4, custom analog audio front-end, and a single Python backend (bridge.py).

AURA listens through an analog microphone, streams audio into a Python backend for STT → AI → TTS, and plays natural speech through a Class-D amplifier.

This repository includes hardware, firmware, Python backend, and 3D-printable shell CAD for a complete end-to-end build.

AURA/
│
├── AURA_ARDUINO/             # PlatformIO firmware project
│   ├── src/main.cpp
│   ├── include/
│   ├── lib/
│   ├── test/
│   ├── .vscode/
│   └── platformio.ini
│
├── bridge.py                 # Single backend Python file (STT → LLM → TTS)
│
├── docs/
│   ├── schematic.pdf         # Full electronics schematic
│   ├── architecture.png      # System architecture diagram
│   └── CAD/
│       ├── AURA Shell UPPER.stl
│       └── AURA Shell LOWER.stl
│
├── resampled_resampy.wav     # Debug sample output
├── README.md
└── .gitignore

AURA uses:

• Arduino UNO R4 Minima
• SEN0487 Analog Electret Microphone
• PAM8302A Class-D Audio Amplifier
• Single-ended mic input (AC-coupled)
• High-frequency PWM speaker output
• Custom 3D-printed enclosure

 Mic → Arduino ADC → Serial → bridge.py →
 Google STT → Azure GPT-4o Mini → TTS →
 Arduino PWM → PAM8302 → Speaker

Additional components used:

• C1 = 10 µF (audio smoothing)
• R1 = 10k (pull-down resistor for stable PWM idle)

docs/schematic.pdf

Located in docs/CAD/:

• AURA Shell UPPER.stl
• AURA Shell LOWER.stl

The Arduino code lives inside AURA_ARDUINO/ and is built using PlatformIO (not Arduino IDE).

AURA_ARDUINO/
│
├── src/main.cpp            # Main firmware logic
├── include/
├── lib/
├── test/
├── .vscode/
└── platformio.ini

main.cpp handles:

• ADC sampling of microphone at ~6.5 kHz
• Serial streaming of raw audio frames to Python
• Receiving PCM audio from Python
• Reconstructing audio via high-frequency PWM on D9
• Synchronizing with Python's bridge.py

From inside AURA_ARDUINO/:

pio run
pio run --target upload

Or use VSCode PlatformIO:

• ✔ Build
• ✔ Upload
• ✔ Serial Monitor

[env:uno_r4_minima]
platform = renesas-ra
board = uno_r4_minima
framework = arduino
monitor_speed = 921600

The entire AI pipeline runs through one Python script.

Audio → STT → GPT-4o Mini → TTS → Audio Playback

• Serial connection to Arduino
• Audio buffer and noise preprocessing
• Google STT
• Azure GPT-4o Mini
• gTTS
• WAV → PCM conversion
• Sending PCM packets to Arduino

git clone https://github.com/l-krrish/AURA
cd AURA

python -m venv .venv
.\.venv\Scripts\activate

pip install -r requirements.txt

AZURE_ENDPOINT = "https://therapy-bot.openai.azure.com/"
AZURE_API_KEY = "<your key>"
AZURE_DEPLOYMENT = "gpt-4o-mini"
GOOGLE_CREDENTIALS = "credentials.json"

Inside AURA_ARDUINO/:

pio run --target upload

python bridge.py

You should see:

Connected to Arduino on COM7
AI Therapy Bot Ready.
Recording...

Speak → AURA replies → audio plays.

• Real-time conversational feedback
• GPT-4o Mini reasoning
• Hardware-accurate PWM audio
• Full schematic + CAD for reproducibility
• Single-file backend for simplicity
• Custom enclosure design

• Wake-word activation
• Noise reduction filters
• ESP32-S3 upgrade
• Local quantized LLM support
• LED expressions + servo motion

Waterloo Computer Engineering

Waterloo Computer Engineering

Waterloo Computer Engineering