Speech Intelligibility in Environmental Sound Maskers and Prediction Based on Envelope-Power Based Models

This study aimed to investigate the speech intelligibility of various environmental sound and to evaluate the three models for predicting speech intelligibility using the observed speech intelligibility in the study. In the experiment, the speech intelligibilities of 28 different environmental sound maskers and two kinds of steady noise maskers were assessed. The three conventional approaches for predicting speech intelligibility, consisting of the sEPSM, the sEPSM correlation-based, and the STOI models, were used to account for speech intelligibility data. The experiments suggested that the envelope power in environmental sound is indicative of speech intelligibility, similar to steady noise. A good correlation between the human data and the corresponding STOI and sEPSMcorr model predictions, based on a correlation metric in the envelope power spectrum domain, were obtained.