ChromaCoverID

Set of functions and methods to compute various chroma and audio similarity measures particularly for the task of cover song identification.

For the moment it includes the python implementation of QMAX, DMAX cover song similarity measures as mentioned in the following papers.

• Serra, J., Serra, X., & Andrzejak, R. G. (2009). Cross recurrence quantification for cover song identification. New Journal of Physics.

• Chen, N., Li, W., & Xiao, H. (2017). Fusing similarity functions for cover song identification. Multimedia Tools and Applications.

[PS: This implementation is purely for research purposes]

Setup

Install python dependencies using pip

$ pip install -r requirements.txt

Usage examples

For more detailed examples have a look on the ipython notebook

• For feature extraction using [chroma_features.py]

from chroma_features import ChromaFeatures

audio_path = "./test_audio.wav"

#Initiate the chroma class
chroma = ChromaFeatures(audio_file=audio_path, mono=True, sample_rate=44100)

# Now you can compute various chroma features and ther plots using the various methods of object chroma
chroma.chroma_stft()
chroma.chroma_cqt()

#You can specify custom params
chroma.chroma_hpcp(hopSize=2048, numBins=24)
chroma.chroma_cens(hopSize=1024)

• Computing cover song similarity measures (qmax and dmax)

from chroma_features import ChromaFeatures
import cover_similarity_measures as sims

chroma1 = ChromaFeatures('<path_to_query_audio_file>')
chroma2 = ChromaFeatures('<path_to_reference_audio_file>')
hpcp1 = chroma1.chroma_hpcp()
hpcp2 = chroma2.chroma_hpcp()

#similarity matrix
cross_recurrent_plot = sims.cross_recurrent_plot(hpcp1, hpcp2)

#cover song similarity distance
qmax, cost_matrix = sims.qmax_measure(cross_recurrent_plot)
dmax, cost_matrix = sims.dmax_measure(cross_recurrent_plot)

Contribute

It would be great if we can compile all the related cover song similarity measures from other papers together in this repo. Let me know if you have any suggestions.

• Fork the repo
• Submit a pull request

Acknowledgements

Thanks to Romain Hennequin for helping in the implementation and Ning Chen for the matlab code for dmax measure.