Performance of modified auditory features for speech signal-an empirical study for non-intrusive speech quality assessment using signal based model
Gitika Sinha, Kumar Shashi Kant · 2018
This paper aims at highlighting improved correlation between objective and subjective speech quality assessment using single ended signal based model. A combination of features like the zero crossing rate, the energy obtained from T eager-Kaiser energy computation, the signal coverage from the Hilbert Transform in form of cepstral coefficients are obtained from the speech signal and compared to classical Mel-frequency cepstral coefficients. Gaussian Mixture Model is used for exploring the features. The correlation between objective and subject MOS on ITU-T P. Supplement 23 database helps in identifying a better technique for quality assessment.