A novel quality measure for time-scale modified speech

Signal-to-noise ratio (SNR) and spectrum distance (SD) are well known objective quality measures for comparing two modified speech frames. However, they are unsuitable for the time-scale modified speech. In this paper, a novel measurement, called spectrum-similarity curve (SSC), is proposed to evaluate the quality of modified speech, especially for the time-scale modified speech. The proposed quality measure is able to perform a local analysis that combined both the time- and spectrum-similarity analyses to achieve the time-similarity plot. The simulation results proved that the sketched curve of the proposed quality measure agrees with the subjective mean opinion score (MOS).