MusicSimilarity-Spark

Feature Extraction Code for Chroma features, Note estimation, MFCC statistics, Rhythm Histogram, Rhythm Patterns and Beat Histograms

Python Spark Code for estimating the similarity of songs

Demo (non-cluster local variant):

• Install dependencies (see below)
• Place audio files in the '''/audio/''' folder and create a target '''/features/''' folder
• Run '''mpiexec -n 4 python example_mass_extractor.py'''
  • Audio features are extracted into the '''/features/''' folder
• Run '''python example_mass_preprocess.py'''
  • A Spark dataframe containing the relevant features is stored to '''AudioFeaturesMerged.json'''
• Run '''spark-submit example_mass_similarity.py'''
  • The first two songs are taken as exemplary song requests

Important Note:

15.Aug.2025: Repository is in the progress of being merged with a repository from a private Gitlab server. Minimal running example for 1 node installation, see above.

Third Party Libraries

Code for Rhythm Histogram and Rhythm Patterns from TU Wien: https://github.com/tuwien-musicir/rp_extract under GNU General Public License v3.0 (slightly adapted version)

Description and Documentation: https://github.com/oObqpdOo/MusicSimilarity

Requirements:

pyspark 1.6.0 or newer,
essentia,
numpy,
scipy,
matplotlib,
urllib,
ipython/ jupyter,
pathlib,
signal, glob, edlib,
ffmpeg or mpg123