Computational and subjective procedures for the assessment of sounds with weak tonal components

Abstract

A discrete tone is defined as ‘‘prominent’’ if the level of the tone exceeds the level of the noise contained in its critical band by 6 dB. Tonal or narrow-band noise components of tire sounds often fall short of this level, but are nevertheless rated as prominent by test drivers. Tonality is closely related to the sensation of pitch strength, which depends upon the sensation level, bandwidth, duration, and frequency of the tonal components present. In order to compare a number of computational procedures for calculating the tonality of sounds, a broad range of tire noises, synthesized tire sounds, and artificial tone–noise mixtures in which the level and bandwidth of the tonal components was varied, were subjectively assessed using (a) a paired-comparison technique, and (b) categorical judgments collected from about 60 listeners. They were objectively assessed by tone-to-noise and prominence-ratio measurements, and by a new method based on a model of pitch sensation which accounts for the temporal and frequency resolution as well as for masking properties of the human ear. The advantages and limits of the different methods will be compared and discussed. Especially for weak narrow-band components, the pitch model seems to be superior to the other methods used to quantify the degree of tonality.