Fabric-sound classification by autoregressive parameters

In order to investigate the sound characteristics of fabrics, a wide range of woven fabrics was selected, and their rustling sounds were recorded. The sounds were analyzed in the forms of sound spectra through Fast Fourier Transform analysis. To evaluate the sound loudness of specimens, LPT (level pressure of total sound) values were calculated. Functions to which the autoregressive (AR) model was applied were used to describe the sound-spectra forms of specimens, and three AR parameters (ARC, ARF, and ARE) of the functions were obtained. Fabric sounds were classified into three clusters by cluster analysis of the parameters. Each of the clusters seemed to be characterized by the parameter ARC, considered to be related to sound loudness, and by the parameter ARE, affecting the over-all shape of the spectrum. To identify the mechanical properties affecting fabric-sound parameters, Kawabata's KES-FB system was used for mechanical measurement. Tensile properties, shear properties, compressional energy, thickness, and weight showed significant differences among the clusters. Of these properties, shear properties seemed to be related to the ARC parameter concerned with sound loudness. Some of the tensile properties and compressional properties were thought to be related to the spectral shapes of fabrics. Finally, shear hysteresis and compressional energy were found to be significantly discriminant for three clusters.