Modulation spectral features for objective voice quality assessment

In this paper, we employ normalized modulation spectral features for objective voice quality assessment regarding grade (hoarseness). Modulation spectra usually produce a high-dimensionality space. For classification purposes, the size of the original space is reduced using Higher Order Singular Value Decomposition (SVD). Further, we select most relevant features based on the mutual information between subjective voice quality (the degree of hoarseness) and the computed features, which leads to an adaptive to the classification task modulation spectral representation. The adaptive modulation spectral features are used as input to a Naive Bayes (NB) classifier. By combining two NB classifiers based on different feature sets a global classification rate of 73.93% for hoarseness was achieved.