Estimating the Properties of the Single-Trial Speech Auditory Brainstem Response Using an Accurate AR Model

The human speech Auditory Brainstem Response (sABR) is an electrophysiological response with potentially important clinical and practical applications. However, because of the very low SNR of the signal, long recording times are usually needed over which the responses from a large number of trials are coherently averaged. Therefore, it is important to understand the properties of the single trial sABR, as this can help in developing methods to detect this response using a smaller number of trials. This paper presents a parametric model of averaged human sABR that is used to estimate the properties of a single-trial response. The Autoregression (AR) method is followed to model the sABR at four different signal qualities, based on recorded data coherently averaged over different numbers of trials. The properties of the modeled sABR are compared with the recorded ones in the time and frequency domains. This model is also used to estimate a single-trial sABR. The results show that the properties of the modeled responses (statistical distribution, SNR, noise power) are similar to the recorded sABRs. Moreover, coherent averaging based on the estimated single-trial sABR produces a comparable theoretical SNR increase, similar exponential relation between the SNR and number of averaged trials, and a similar power spectrum to the recorded sABR.