The Algorithm And Realization Of Psychoacoustic Metrics

The algorithm of psychoacoustic metrics is a method based on computational procedure for sound quality evaluation. Compared with the method of jury listening test, it's repeatable, more efficient and conducive to standardization of sound quality evaluation. Psychoacoustic metrics include loudness, sharpness, fluctuation strength, roughness, language interference level and the clarity index and so on. Each metric has serval different computational models. For the loudness, there are Stevens model, Zwicker model and Moore model. For the sharpness, there are Aures model and Von Bismarck model. At present, only loudness has been standardized and has international standards. These make the application of psychoacoustic metrics a little difficult in the real world.To decrease the heterogeneity of psychoacoustic metrics, this thesis emphasizes on the programming of several models for loudness, sharpness as well as roughness, and compares with each other using theoretical analysis and subject test. Therefore, people can pick out an appropriate model when the sound quality evaluation is in need.In the first place, the superiority of Moore model is apparent, through the analysis and comparison of three standard algorithms for loudness calculation including Stevens model, Zwicker model and Moore model. After analyzing the deficiency of the three accessible input methods of Moore model (1/3 octave spectra, FFT spectrum, synthetic spectra of several FFTs spectrum), a new approach is proposed to improve the efficiency and precision, which uses ERB scale to divide the auditory frequency into a series of bands, and demonstrates effectiveness according to computational results.In the second place, several algorithms for sharpness calculation, including Zwicker model, Aures model and Von Bismarck model etc, are compared, of which all the models are based on the Zwicker's specific loudness. Due to the difference of distribution in Bark scale between Moore's and Zwicker's specific loudness, Moore's specific loudness cannot be used to calculate the sharpness directly. In addition, the thesis turns to the analysis of the principle of Aures roughness model and Vassilakis roughness model, and the in-depth study of implementation procedure of Aures roughness model, which covers some details leading to errors easily as well. The calculation of AM tones'roughness has demonstrated the limitations of Vassilakis model. Additionally, the thesis outlines the Terhardt pitch extraction procedure and realizes the process through programming.Finally, several subjective evaluation tests are carried out using magnitude estimation method. Loudness, sharpness and roughness of routine noise are evaluated. It is demonstrated that three standard loudness models are in good agreement with subjective sensation; there is some discrepancy between the existing sharpness models and subjective sensation; Aures roughness model can't reflect people's subjective feeling sometimes.