Quantifying Sound Quality in Loudspeaker Reproduction

Abstract

We present an objective perceptual measurement method for the assessment of the perceived sound quality of loudspeakers, based on the core elements found in the perceptual evaluation models as developed within ITU. Instead of quantifying the loudspeaker system itself, the model quantifies the overall perceived sound quality of loudspeakers by assessing their acoustic output. This approach introduces a major problem. We cannot provide an acoustic reference signal to the subject that can be directly compared to the acoustic degraded loudspeaker output. A solution for this problem is proposed by creating binaural recordings of the reproduced reference signal with a Head and Torso Simulator (HATS), using the best quality loudspeakers available, in the ideal listening spot in the best quality listening environment available. The reproduced reference signal with the highest subjective quality given by subjects is compared to the acoustic degraded loudspeaker output. The model is developed using three large databases that contain binaural recorded music fragments played over very low, to very high quality loudspeakers in very low, to very high quality listening rooms. The average error in percentage between the training (r = 0.90) and the validation of the model (r = 0.85) is 5.6%, showing the high stability of the model. As such, the model is a suitable candidate to accurately quantify the sound quality in loudspeaker reproduction.