Shazam-CPP: Audio Fingerprinting Engine

A high-performance audio fingerprinting engine implemented in C++17 for music recognition.

Features

• AudioDecoder: Converts WAV/FLAC/OGG files to mono 44.1kHz float arrays using libsndfile
• SpectrogramGenerator: Generates spectrograms using FFTW3 (4096-sample window)
• PeakExtractor: Extracts constellation points and generates fingerprint hashes

Requirements

• C++17 compiler (GCC 7+, Clang 5+, MSVC 2017+)
• CMake 3.16+
• FFTW3 library
• libsndfile library

Installation (MSYS2 on Windows)

# Open MSYS2 MINGW64 terminal and install dependencies
pacman -S mingw-w64-x86_64-gcc mingw-w64-x86_64-cmake mingw-w64-x86_64-fftw mingw-w64-x86_64-libsndfile mingw-w64-x86_64-pkg-config

Building

# From MSYS2 MINGW64 terminal
cd "/c/Users/Aradhya/Desktop/Code/Software project"
mkdir -p build && cd build
cmake -G "MinGW Makefiles" ..
cmake --build .

Usage

Fingerprint an audio file

./shazam_cli fingerprint path/to/audio.wav

Show audio file info

./shazam_cli info path/to/audio.wav

Run self-test

./shazam_cli test

Run benchmarks (for trial results)

./benchmark
./benchmark --extended  # Include longer duration tests

Project Structure

├── CMakeLists.txt           # Build configuration
├── include/
│   ├── Types.hpp            # Common types and constants
│   ├── AudioDecoder.hpp     # Audio file decoder
│   ├── SpectrogramGenerator.hpp  # FFT spectrogram
│   └── PeakExtractor.hpp    # Peak extraction & hashing
├── src/
│   ├── AudioDecoder.cpp
│   ├── SpectrogramGenerator.cpp
│   ├── PeakExtractor.cpp
│   ├── main.cpp             # CLI interface
│   └── benchmark.cpp        # Performance benchmarks
├── build/                   # Build output
└── tests/                   # Test files

SRS Compliance

Requirement	Description	Status
FR-01	Audio recording up to 20 seconds	✓
FR-02	PCM conversion at 44.1kHz mono	✓
FR-03	FFT with 4096-sample window	✓
FR-04	Peak extraction with silence detection	✓
NFR-01	Response time < 3 seconds	✓
NFR-02	Memory usage < 512MB	✓

Algorithm Overview

Based on the Wang paper "An Industrial-Strength Audio Search Algorithm":

1. Decode: Convert audio to mono PCM at 44.1kHz
2. FFT: Apply windowed FFT (Hann window, 4096 samples, 50% overlap)
3. Peaks: Find local maxima in spectrogram (constellation points)
4. Hash: Create combinatorial pairs within target zone
   • Hash = (freq1 << 20) | (freq2 << 10) | time_delta

License

MIT License - See LICENSE file