Perceptual Speech Quality Measure (PSQM)

PSQM was developed by PTT Research in 1994 [15,14]. It can be considered as a modified version of the Perceptual Audio Quality Measure (PAQM) which is an objective audio quality measure also developed at PTT Research [14]. Recognizing that the characteristics of speech and music are different, PSQM was optimized for speech by modifying some of the procedures of PAQM. PSQM has been adopted as ITU-T Recommendation P.861 [71]. Its performance has been shown to be relatively robust for coding distortions. PSQM transforms the speech signal into the loudness domain, modifying some parameters in the loudness calculation in order to optimize performance. PSQM does not include temporal or spectral masking in its calculation of loudness. It applies a nonlinear scaling factor to the loudness vector of distorted speech. The scaling factor is obtained using the loudness ratio of the reference and the distorted speech in three frequency bands. The difference between the scaled loudness of the distorted speech and the loudness of the reference speech is called noise disturbance. The final estimated distortion is an averaged noise disturbance over all the frames processed. PSQM disregards or applies a small weight to silence portions in the calculation of distortion. PSQM uses psychoacoustic results of loudness calculation to transform speech into the perceptually relevant domain. It also considers the role of distortions in silence portions on overall speech quality. Even though its performance is relatively robust over coding distortions, it may not be robust enough to apply to a broader range of distortions.