Perceptual Quality Sound Separation

Evaluation of Quality of Sound Source Separation Algorithms: Human Perception vs Quantitative Metrics

Authors: Estefanía Cano, Derry Fitzgerald, Karlheinz Brandenburg

This is the supplementary material for the work presented in EUSIPCO 2016.


In this study, we look into the test methods to evaluate the quality of audio separation algorithms. Specifically, we try to correlate the results of listening tests with state-of-the-art separation quality measures. To this end, the quality of the harmonic signals obtained with two harmonic-percussive separation algorithms was evaluated with BSS, PEASS and via listening tests. Results show that for harmonic-percussive separation algorithms, neither BSS nor PEASS show strong correlation with the ratings obtained via listening tests and suggest that perceptual objective measures for quality assessment do not generalize well to different separation algorithms.


A dataset composed of 10 music signals was used for this experiment. Audio mixtures for all signals were obtained using the original multi-track recordings and were processed with the two HP algorithms. In this study, only the extracted harmonic signals were used in the evaluation.


In this study, the quality of two separation algorithms, alg1 [1] and alg2 [2], was evaluated under the common dataset.

[1] Estefanía Cano, Mark Plumbley, and Christian Dittmar, “Phase-based harmonic/percussive separation,” in 15th Annual Conference of the International Speech Communication Association (2014). Interspeech., 2014.

[2] Derry FitzGerald, Antoine Liutkus, Zafar Rafii, Bryan Pardo, and Laurent Daudet, “Harmonic/percussive separation using kernel additive modelling,” in Proceedings of the Irish Signals and Systems Conference, 2014.