Jose/esp32i2s
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
c83d04eb23d643860cb3dfd7b3f274bb82b3f25a
esp32i2s/README.md
Jose c83d04eb23 plaintext feat undefined: Updated Arduino project with new features for audio processing. fix: Updated I2SConfig.h to include additional parameter for reading I2S samples. refactor: Improved performance by rounding integer samples to 
- This code snippet is a C++ program that utilizes the Arduino framework to implement audio processing for a microphone. The program includes several classes and functions to handle various aspects of audio analysis, such as note detection, frequency analysis, and spectrum visualization.

The key components of this program include:

1. **AudioLevelTracker**: This class provides real-time audio level monitoring by tracking the maximum amplitude of an input signal. It uses a simple peak detection algorithm to determine when the input signal reaches a certain threshold.

2. **NoteDetector**: This class performs frequency analysis on the input audio signal and identifies specific notes based on their frequencies. The note detector utilizes a pre-defined list of known frequencies and compares them against the detected frequencies to identify matches.

3. **SpectrumVisualizer**: This class provides real-time spectrum visualization by displaying the magnitude of the input audio signal in the form of an ASCII graph. The magnitude scaling is done dynamically based on the signal power to ensure that all frequencies are visible.

4. **Main Loop**: The main loop handles all the other components and processes them sequentially. It initializes the audio level tracker, note detector, and spectrum visualizer, and then enters a loop where it continuously processes the input audio signal.

The program also includes error handling mechanisms, such as automatic I2S reset on communication errors and dynamic threshold adjustment to ensure that the audio processing remains stable and accurate. The project is structured with clear class definitions and proper documentation for each component.
- The updateMaxLevel and getMaxLevel methods in AudioLevelTracker have been modified to accept and return int16_t values instead of int32_t, which improves range handling.
- The `Config.h` file has been updated to enhance audio processing by increasing gain, adjusting noise threshold for 16-bit samples, and changing the FFT size from a power of 2. The main goal is to optimize performance while maintaining good noise detection and note detection capabilities for better accuracy in music analysis tasks.
- The `git diff` output shows a change to the I2SConfig.h file. Specifically, it adds a line to define an additional parameter for reading I2S samples: int16_t*.
- This commit introduces a new header file `NoteDetector.h` for detecting musical notes in an Arduino project, enhancing the detection process with FFT analysis and dynamic threshold adjustments.
- The `SpectrumVisualizer.h` file has been added to the project with new definitions and functions to visualize audio spectrum and detected notes.
- The main goal of these changes is to update the `lib_deps` in the `platformio.ini` file to include a specific library named `kosme/arduinoFFT` which is version 1.6.
- The changes improve the audio level tracking by rounding the integer samples to 16 bits before storing them, ensuring that the range remains within a feasible limit for processing.
- The main goal of the changes is to optimize the `readI2SSamples` function by removing unnecessary conversion from 16-bit to 32-bit samples, which was previously done in an existing code section that could be reused for other purposes. This change improves performance and reduces complexity while maintaining compatibility with existing code.
- A new `NoteDetector` class has been created in the `src/NoteDetector.cpp` file, implementing various calibration and note detection functionalities.
- The user has added new functions `magnitudeToDb`, `mapToDisplay`, `printBarGraph`, `drawFFTMagnitudes`, `visualizeSpectrum`, and `visualizeNotes` to the `SpectrumVisualizer.cpp` file. The changes are related to visualizing spectrum data and note detection results in a serial monitor format for debugging.
- The main goal is to enhance the piano note detection system by adding support for a NoteDetector and updating SpectrumVisualizer when notes are detected, as well as handling serial commands for calibration, threshold adjustments, and toggling spectrum display.
2025-04-25 12:14:06 +02:00

6.8 KiB
Raw Blame History

ESP32 Piano Note Detection System

A real-time piano note detection system implemented on ESP32 using I2S microphone input. This system can detect musical notes from C2 to C6 with adjustable sensitivity and visualization options.

Features

  • Real-time audio processing using I2S microphone
  • FFT-based frequency analysis
  • Note detection from C2 (65.41 Hz) to C6 (1046.50 Hz)
  • Dynamic threshold calibration
  • Multiple note detection (up to 7 simultaneous notes)
  • Harmonic filtering
  • Real-time spectrum visualization
  • Note timing and duration tracking
  • Interactive Serial commands for system tuning

Hardware Requirements

  • ESP32 development board
  • I2S MEMS microphone (e.g., INMP441, SPH0645)
  • USB connection for Serial monitoring

Pin Configuration

The system uses the following I2S pins by default (configurable in Config.h):

  • SCK (Serial Clock): GPIO 8
  • WS/LRC (Word Select/Left-Right Clock): GPIO 9
  • SD (Serial Data): GPIO 10

Getting Started

  1. Connect the I2S microphone to the ESP32 according to the pin configuration
  2. Build and flash the project to your ESP32
  3. Open a Serial monitor at 115200 baud
  4. Follow the calibration process on first run

Serial Commands

The system can be controlled via Serial commands:

  • h - Display help menu
  • c - Start calibration process
  • + - Increase sensitivity (threshold up)
  • - - Decrease sensitivity (threshold down)
  • s - Toggle spectrum visualization

Configuration Options

All system parameters can be adjusted in Config.h:

Audio Processing

  • SAMPLE_RATE: 8000 Hz (good for frequencies up to 4kHz)
  • BITS_PER_SAMPLE: 16-bit resolution
  • SAMPLE_BUFFER_SIZE: 1024 samples
  • FFT_SIZE: 1024 points

Note Detection

  • NOTE_FREQ_C2: 65.41 Hz (lowest detectable note)
  • NOTE_FREQ_C6: 1046.50 Hz (highest detectable note)
  • FREQUENCY_TOLERANCE: 3.0 Hz
  • MAX_SIMULTANEOUS_NOTES: 7
  • MIN_NOTE_DURATION_MS: 50ms
  • NOTE_RELEASE_TIME_MS: 100ms

Calibration

  • CALIBRATION_DURATION_MS: 5000ms
  • CALIBRATION_PEAK_PERCENTILE: 0.95 (95th percentile)

Visualization

The system provides two visualization modes:

  1. Note Display:
Current Notes:
A4 (440.0 Hz, Magnitude: 2500, Duration: 250ms)
E5 (659.3 Hz, Magnitude: 1800, Duration: 150ms)
  1. Spectrum Display (when enabled):
Frequency Spectrum:
0Hz    |▄▄▄▄▄
100Hz  |██████▄
200Hz  |▄▄▄
...

Performance Tuning

  1. Start with calibration by pressing 'c' in a quiet environment
  2. Play notes and observe the detection accuracy
  3. Use '+' and '-' to adjust sensitivity if needed
  4. Enable spectrum display with 's' to visualize frequency content
  5. Adjust Config.h parameters if needed for your specific setup

Implementation Details

  • Uses FFT for frequency analysis
  • Implements peak detection with dynamic thresholding
  • Filters out harmonics to prevent duplicate detections
  • Tracks note timing and duration
  • Uses ring buffer for real-time processing
  • Calibration collects ambient noise profile

Troubleshooting

  1. No notes detected:

    • Check microphone connection
    • Run calibration
    • Increase sensitivity with '+'
    • Verify audio input level in spectrum display

  2. False detections:

    • Run calibration in a quiet environment
    • Decrease sensitivity with '-'
    • Adjust PEAK_RATIO_THRESHOLD in Config.h

  3. Missing notes:

    • Check if notes are within C2-C6 range
    • Increase FREQUENCY_TOLERANCE
    • Decrease MIN_MAGNITUDE_THRESHOLD

Contributing

Contributions are welcome! Please read the contributing guidelines before submitting pull requests.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Development Environment Setup

Prerequisites

  • PlatformIO IDE (recommended) or Arduino IDE
  • ESP32 board support package
  • Required libraries:
    • arduino-audio-tools
    • arduino-audio-driver
    • WiFiManager
    • AsyncTCP
    • ESPAsyncWebServer
    • arduinoFFT

Building with PlatformIO

  1. Clone the repository
  2. Open the project in PlatformIO
  3. Install dependencies: 
    pio lib install
  4. Build and upload: 
    pio run -t upload

Memory Management

Memory Usage

  • Program Memory: ~800KB
  • RAM Usage: ~100KB
  • DMA Buffers: 4 x 512 bytes
  • FFT Working Buffer: 2048 bytes (1024 samples x 2 bytes)

Optimization Tips

  • Adjust DMA_BUFFER_COUNT based on available RAM
  • Reduce SAMPLE_BUFFER_SIZE for lower latency
  • Use PSRAM if available for larger buffer sizes

Advanced Configuration

Task Management

  • Audio processing runs on Core 1
  • Main loop on Core 0
  • Configurable priorities in Config.h

Audio Pipeline

  1. I2S DMA Input
  2. Sample Buffer Collection
  3. FFT Processing
  4. Peak Detection
  5. Note Identification
  6. Output Generation

Timing Parameters

  • Audio Buffer Processing: ~8ms
  • FFT Computation: ~5ms
  • Note Detection: ~2ms
  • Total Latency: ~15-20ms

Performance Optimization

CPU Usage

  • Audio Processing: ~30% on Core 1
  • Note Detection: ~20% on Core 1
  • Visualization: ~10% on Core 0

Memory Optimization

  1. Buffer Size Selection:
    • Larger buffers: Better frequency resolution
    • Smaller buffers: Lower latency
  2. DMA Configuration:
    • More buffers: Better continuity
    • Fewer buffers: Lower memory usage

Frequency Analysis

  • FFT Resolution: 7.8125 Hz (8000/1024)
  • Frequency Bins: 512 (Nyquist limit)
  • Useful Range: 65.41 Hz to 1046.50 Hz
  • Window Function: Hamming

Technical Details

Microphone Specifications

  • Supply Voltage: 3.3V
  • Sampling Rate: 8kHz
  • Bit Depth: 16-bit
  • SNR: >65dB (typical)

Signal Processing

  1. Pre-processing:
    • DC offset removal
    • Windowing function application
  2. FFT Processing:
    • 1024-point real FFT
    • Magnitude calculation
  3. Post-processing:
    • Peak detection
    • Harmonic filtering
    • Note matching

Calibration Process

  1. Ambient Noise Collection (5 seconds)
  2. Frequency Bin Analysis
  3. Threshold Calculation:
    • Base threshold from 95th percentile
    • Per-bin noise floor mapping
  4. Dynamic Adjustment

Error Handling

Common Issues

  1. I2S Communication Errors:
    • Check pin connections
    • Verify I2S configuration
    • Monitor serial output for error codes
  2. Memory Issues:
    • Watch heap fragmentation
    • Monitor stack usage
    • Check DMA buffer allocation

Error Recovery

  • Automatic I2S reset on communication errors
  • Dynamic threshold adjustment
  • Watchdog timer protection

Project Structure

Core Components

  1. AudioLevelTracker
    • Real-time audio level monitoring
    • Peak detection
    • Threshold management
  2. NoteDetector
    • Frequency analysis
    • Note identification
    • Harmonic filtering
  3. SpectrumVisualizer
    • Real-time spectrum display
    • Magnitude scaling
    • ASCII visualization

File Organization

  • /src: Core implementation files
  • /include: Header files and configurations
  • /data: Additional resources
  • /test: Unit tests
Reference in New Issue View Git Blame Copy Permalink