Neural Acoustic-Phonetic Approach for Speaker Verification With Phonetic Attention Mask

Traditional acoustic-phonetic approach makes use of both spectral and phonetic information when comparing the voice speakers. While units are not equally informative, context speech plays an important role in speaker verification (SV). In this paper, we propose a neural that learns to dynamically assign differentiated weights features for SV. Such form attention mask (PAM). The framework consists two training pipelines, one SV another recognition. Through PAM, leverage We evaluate proposed on RSR2015 database Part III corpus, random digit strings. show with PAM consistently outperforms baseline equal error rate reduction 13.45% 10.20% female male data, respectively.