🎵 Song Recognition Algorithm (Audio Fingerprinting)

A Python implementation of a Shazam-like song recognition system based on spectrogram peak constellation maps and audio fingerprint hashing.
The system identifies a song from a short audio clip by matching time-aligned spectral fingerprints against a database.

---

📌 Overview

This project demonstrates how audio fingerprinting works at a low level using:

• Digital Signal Processing (DSP)
• Spectrogram peak detection
• Hash-based matching with temporal alignment

Given:

• A database of songs
• A short query clip (e.g., 10 seconds)

The system returns:

• The most likely matching song
• A confidence score based on aligned hash matches

---

🧠 Core Idea (How It Works)

1. Convert audio to mono and normalize
2. Compute spectrogram (STFT)
3. Detect prominent spectral peaks
4. Create a constellation map
5. Generate compact fingerprints (hashes)
6. Match query fingerprints against database
7. Vote using time-offset alignment

This approach is inspired by Shazam’s audio fingerprinting technique.

---

🧩 File Breakdown

constellation.py

Contains all DSP and fingerprinting logic.

Key Functions

• read_wav_file()
  Reads WAV audio, converts stereo to mono, and normalizes amplitude.

• compute_spectrogram()
  Computes magnitude spectrogram using FFT and Hann windowing.

• detect_spectrogram_peaks()
  Detects local maxima using neighborhood filtering and thresholding.

• generate_fingerprints()
  Generates compact hashes from anchor–target peak pairs using frequency and time deltas.

---

testing.py

Implements database creation and song matching.

Database Structure

{
  "Song_name": "...",
  "Fingerprint": [(hash, time_offset), ...]
}