MotivesExtractor

This script extracts the most repeated harmonic patterns from an audio file sampled at 11025Hz. It is based on the following paper:

Nieto, O., Farbood, M., Identifying Polyphonic Patterns From Audio Recordings Using Music Segmentation Techniques. Proc. of the 15th International Society for Music Information Retrieval Conference (ISMIR). Taipei, Taiwan, 2014.

Examples

To run the extractor on a single file without CSV annotations (the results will be printed on the screen):

./extractor.py wav_file

where wav_file is the path to a wav file sampled at 11025Hz with 16 bits. You can find the wav files from the JKU dataset in the folder jku_input.

To run the extractor on a single file with CSV annotations:

./extractor.py wav_file -c csv_file [-o output.txt]

where csv_file is the path to the corresponsing CSV file using the JKU format. Examples of CSV files are included in the jku_input folder. The output is saved using the MIREX format. If the output file is not provided, the results will be saved in results.txt.

To run the extractor on multiple files:

./run_extractor.py input_folder output_folder [-j 8]

It will analyze all the wav files and their corresponding CSV files from the input_folder and write the results into the output_folder. The parameter -j indicates how many processors you want to run in parallel (default is 4).

To evaluate an entire folder:

./eval.py references_folder estimations_folder

where these folders contain the patterns using the MIREX format, and both folders contains the exact same file names.

To obtain the ISMIR numbers:

./run_extractor.py jku_input/ results/ -th .33 -r 2
./eval.py parsed_jku/ results/

To plot the ISMIR paper plots:

./extractor.py jku_input/mazurka24-4-poly.wav -th .33 -r 2 -ismir

For more options, please type

./run_extractor.py -h

Requirements

• Python >=2.7
• audiolab
• numpy
• scipy
• joblib
• pylab (For plotting only)
• pandas (For evaluating only)
• mir_eval (For evaluating only)

Author

Oriol Nieto