Stimulus‐based diotic and dichotic models that combined cues for detection of tones in reproducible noise.

Abstract

Understanding speech in noisy backgrounds is a classical problem for people with hearing loss. The mechanisms for detection of pure tones in noise are still not fully understood. In this study, models that combined cues across epochs were constructed to predict binaural psychophysical experimental results. The data came from previous experiments, in which listeners were required to detect a 500‐Hz tone in either narrowband or wideband reproducible noise waveforms in both diotic and dichotic conditions. For the diotic condition, cues were correlated and could be described by a multivariate distribution. Cues were combined using the covariance matrix. Decision variables based on the likelihood ratio of the tone presence were compared to the listeners’ detection patterns. Energy, envelope‐slope, and fine‐structure cues were combined optimally in the model. For the dichotic condition, the same strategy was applied by combining interaural level and interaural time differences. Better predictions were obtained from this model compared with models that relied on a weighted sum of the two cues. In addition, a modified envelope‐slope model that computed the slope of the difference of the envelopes at the two ears yielded better predictions than models based on interaural time and level differences for the dichotic condition. [NIH‐NIDCD‐R01DCO1O813.]