Audio identification is a class of information retrieval systems designed to match a query with a set of database documents. The objective is to find the most similar track in a database given a short audio snippet.
This implementation utilizes Wang's fingerprinting method (Wang 2003), leveraging frequency pairs and time difference hashes to efficiently match query audio with the database.
- 80% Recall at Rank 1
- 90% Recall at Rank 3
- Dataset: GTZAN Music Dataset
The following Python libraries are required:
- Numpy
- Librosa
- Matplotlib
- Skimage
Install the required packages via pip:
pip install numpy librosa matplotlib scikit-image- An Industrial-Strength Audio Search Algorithm (Wang, 2003)