Contribution
Modeling Binaural Speech Intelligibility in Listening Conditions with Spectro-Temporally Modulated Maskers
* Presenting author
Abstract:
The binaural speech-based envelope power spectrum model is based on the analysis of amplitude modulations in the noise signal and the noisy target speech, combined with a better-ear (BE) and binaural unmasking (BU) process. While the model was shown to accurately represent speech perception in asymmetric-interferer conditions, its performance in symmetric setups remains unclear. In the model, BE and BU are considered in the modulation domain using short-time processing, thus, it was hypothesized that the model should also account for the spatial effects observed in symmetric masker conditions with spectro-temporally modulated maskers. In the present study, the model was evaluated using an existing dataset that systematically investigated the dependencies of speech reception thresholds and spatial release from masking (SRM) on spectro-temporal masker properties in symmetric, anechoic 2-masker conditions. The maskers considered for modelling were stationary speech-shaped noise (SSN), sinusoidal amplitude-modulated SSN, speech-modulated SSN and SSN speech-modulated incoherently across-frequency. The model’s binaural processing was assessed using alternate model versions, considering only BE or BU, to predict the SRM in conditions where only interaural level differences or interaural time differences were available. Preliminary results show that BE and BU in the modulation domain may reliably predict SRM across a variety of scenarios.