Research On Objective Measures Of Speech Quality Based On Masking Properties Of Auditory System

Speech signal is an important means in communication. Currently as voice communication systems have been rapidly change, there is increasing interest in the development of a speech quality assessment. There are two ways include subjective and objective speech quality of assessment in principle. The most widely-used objective speech quality measure demonstrate the performance of speech codecs is the in-output-based objective measure of speech quality, based on the measure of spectral distortion. This dissertation focuses on the researches of modified Bark spectral distortion measure with the auditory masking properties, as well as their application to objective assessments of speech quality. The major contributions of this dissertation are as follows:1.The BSD measure is based on a perceptual model that incorporates the frequency smearing within a given short-time frame, but not taken into account the inter-frame masking effects that impact human perceptual discrimination. Hence an opportunity exists to incorporate these inter-frame temporal masking effects into the BSD measure to improve its performance. Is computed only from perceptual distortion of the speech in the MBSD measure, imperceptible distance contributions are ignored. The MBSD defines the distortion as the average difference of estimated loudness. Preliminary experimental results show that the MBSD measure is a preferable speech quality assessment that can consist with subjective assessment of speech quality well and the correlative coefficient of MBSD over the conventional BSD.2.A psychoacoustic experiment shows that human hearing masking effects is not only temporal masking, is but also including simultaneous masking. Another part of this dissertation studies the EBSD measure with simultaneous masking. It can embody adequately the perception properties of the human auditory system with simple computing. It is found in experimentalinvestigation that the correlative coefficient of the MOS predicted by EBSD and the subjective MOS measure can reach up to about 0.95. Noise masking threshold plays an important role in estimating perceptual distortion in the EBSD. The performance of the EBSD measure has been examined for speech data with coding distortions by varying the scaling factor. Compared with the MBSD, The performance of the EBSD measure is slightly better than that of the MBSD.We can conclude that these measures have certain validity and utility to appraise the quality of speech systems. Objective distortion measure with perceptual properties is studied deep, which has been applied to speech quality assessment successfully. Not only in militarily, but also is widely applied to research of feature extraction in speech enhancement and speech recognition.