Estimation of the critical bandwidth from loudness summation data.

An automatic method for critical band estimation from loudness summation data is presented. A mathematical model, based on a power function, is fitted to the data and the critical bandwidth is defined at the intersection of the asymptotes. The model is designed for clinical use, involving the treatment of the data from single test persons; it represents an operational solution to a difficult task. The model is able to describe data from normals and patients with a sensorineural hearing loss. Variability of the critical band estimates, intrasubject as well as intersubject, is larger than for visually obtained estimates. However, visual estimation is difficult, subjective, and probably heavily biased. The model produces estimates which in logarithmic form have a Normal distribution at medium loudness level, while visual estimation gives rise to irregular distributions at all levels. A normal range for model estimates from sets of data obtained at medium loudness level is defined by mean and standard deviation.