Speech Intelligibility Prediction Based on the Envelope Power Spectrum Model with the Dynamic Compressive Gammachirp Auditory Filterbank

In this study, we develop a new method to realize speech intelligibility prediction of synthetic sounds processed by nonlinear speech enhancement algorithms. A speech envelope power spectrum model (sEPSM) was proposed to account for subjective results on a spectral subtraction, but it is untested by recent state-of-the-art speech enhancement algorithms. We introduce a dynamic compressive gammachirp auditory filterbank as the front-end of the sEPSM (dcGC-sEPSM) to improve the predictability. We perform subjective experiments on speech intelligibility (SI) of noise-reduced sounds processed by the spectral subtraction and a recently developed Wiener filter algorithm. We compare the subjective SI scores with the objective SI scores predicted by the proposed dcGC-sEPSM, the original GT-sEPSM, the three-level coherence SII (CSII), and the short-time objective intelligibility (STOI). The results show that the proposed dcGC-sEPSM performs better than the conventional models.