Selecting Program Material by Audio Features for Low-Frequency Perceptual Evaluation of Loudspeakers

Abstract

The program material is of great importance for the results of the listening tests on loudspeakers, while the process of how to select the program material remains ambiguous. This paper investigates the criterion for selecting programs suitable for low-frequency perceptual evaluation based on the audio features of the program. A listening test was conducted to identify the more discriminating and revealing programs in the low-frequency range. Based on the listening test results, various characteristics of the programs, including dynamic, timbral, rhythmic, and spectral features, were extracted. Their relationships with the program’s discrimination ability are discussed. The results suggest that programs with a slow and clear rhythm and a smooth and even spectrum in the whole band are more discriminating in detecting the spectral differences in the low frequencies. By using these significant features, a discriminant analysis was performed to predict the ability of the program to reveal the spectral irregularities. The predictive accuracy of the derived discriminant function was 95% in separating the discriminating and undiscriminating programs.