CASA-Based Robust Speaker Verification Using Popular Features in Real Environments

In this paper, the ideal binary mask is implemented in accordance with popular features: GFCC, PNCC, MFCC, and RASTAPLP to apply in speaker verification system. Inspired by auditory perception, computational auditory scene analysis the ideal binary mask is used to eliminate the combined effects of reverberation and noise. The performance of the CASA-based SV systems are evaluated using the EER and DET curve and compared with the conventional SV systems. Evaluation on the IEEE corpus shows that the CASA-based SV systems substantially, outperform the conventional SV in both of reverberation and reverberation + noise conditions and among the CASA-based SV system the PN-IBM-R SV has the best performance.

Keywords:

Ideal binary mask , reverberation and noise , robust speaker verification