Using Psychoacoustic Models for Sound Analysis in Music

Abstract

Overall sound perception of a song is an important attribute of music. Several psychoacoustic models have been studied to extract perceptual sound qualities from audio signals. By means of listening tests, we investigate whether these sound models successfully reflect (inter-)subjective perception of sound resemblance in music. Preliminary results shows that psychoacoustic descriptors are in better accordance with subjective judgments than low-level features. As the psychoacoustic descriptors model auditory perception, we can assume causal relationships and directly draw conclusions on perception from the listening test results. We observed that roughness is the most crucial sound resemblance criteria, followed by sharpness, loudness, spaciousness and tonalness. The findings indicate that these psychoacoustic models may be suitable for music data mining, music browsing, automatic playlist generation and music recommendation tasks.