🎙️ ESP32 Audio Data Collector

A high-speed Serial utility for building machine learning audio datasets. This tool allows you to stream raw 16-bit PCM data from an ESP32, visualize signal levels in real-time, and save recordings into labeled .wav files.

🚀 Features

• Live Debug Mode: A real-time ASCII volume meter to verify microphone gain and wiring without saving files.
• Bulk Recording: Optimized workflow for capturing dozens of samples quickly—perfect for "Wake Word" training.
• High-Speed Transfer: Uses 921,600 Baud to ensure 16kHz audio data is captured without packet loss.
• Automatic Organization: Files are timestamped and sorted by label (e.g., command_170921400_1.wav).

---

🛠️ Requirements

Hardware

• ESP32 (S3, C3, or Standard DevKit).
• I2S Microphone (e.g., INMP441, ICS-43434) or an Analog Mic using the ESP32 ADC.
• USB Data Cable (Ensure it is a data cable, not just a charging cable).

Software

• Python 3.7+
• PySerial library:

pip install pyserial

---

📂 Installation & Setup

1. Clone the Repository:

git clone https://github.com/jhoward-embedded/EdgeImpulseDataRecorder-ESP32.git
cd EdgeImpulseDataRecorder-ESP32

2. Configure the Script: Open audio_collector.py and update the PORT variable to match your ESP32:

PORT = 'COM5'          # Windows
# PORT = '/dev/ttyUSB0' # Linux/Mac

3. ESP32 Firmware: Ensure your ESP32 is programmed to:

• Send raw binary (2 bytes per sample) when it receives the character 's'.
• Send ASCII integers (text lines) when it receives the character 'd'.

---

📖 How to Use

Run the script:

python audio_recorder.py

1. Live Debug Mode

Choose Option 1 to test your microphone. You will see a level meter like this: [########## ] 4200 If the bar doesn't move when you speak, check your I2S clock (BCLK/WS) and Data (SD) wiring.

2. Data Collection

Choose Option 2 to start building your dataset.

1. Enter a label when prompted (e.g., alexa, background, noise).
2. Press Enter to start a 3-second capture.
3. The script will automatically increment the file count and save it to the dataset_recordings/ folder.
4. Type q to stop the batch and return to the main menu.

---

📉 Technical Specs

Parameter	Value
Sample Rate	16,000 Hz
Bit Depth	16-bit Signed Integer (PCM)
Baud Rate	921,600 bps
Format	Mono .WAV (Little Endian)

---

🤝 Contributing

Contributions are welcome! If you'd like to add features like a Spectrogram preview or automatic silence trimming, please open an issue or a Pull Request.

---