Audio Fingerprinting — EE200 Project Demo

Shazam-style audio fingerprinting app built for the EE200 (Signals, Systems & Networks) course project.

Setup

1. Install dependencies:

   pip install -r requirements.txt

2. Build the fingerprint database (run once — takes ~2 min):

   python build_db.py

3. Generate sample query clips (for the Identify tab):

   python generate_samples.py

4. Launch the app:

   streamlit run app.py

How It Works

1. Spectrogram — each song is turned into a time-frequency image using the STFT.
2. Constellation — local maxima (peaks) in the spectrogram are extracted as landmark points.
3. Hashing — nearby peaks are paired into compact hashes (f1, f2, Δt) for efficient lookup.
4. Matching — query hashes are looked up in the pre-built database; matching hashes vote for a time offset, and a true match produces a sharp spike in the offset histogram.

App Tabs

Tab	What it does
Library	Lists all indexed songs with duration and fingerprint stats
Identify	Upload a clip (or pick a sample) → see the full pipeline: spectrogram, constellation, offset histogram, matched song
Batch	Upload multiple clips → get a results.csv with filename, prediction columns

File Structure

File	Purpose
fingerprint.py	Core fingerprinting engine
build_db.py	Indexes all songs into pickle databases
generate_samples.py	Creates sample query clips
app.py	Streamlit web app
Q3_data/	Song library (50 .mp3 files)
samples/	Sample query clips
song_database.pkl	Pre-built hash database
song_metadata.pkl	Song metadata (durations, peaks)