Speaker verification (SV) is the task of validating the identity claim of a person. The existing SV systems model speaker information by including features from all the categories of sound units together. There is large variability among different sound units and may benefit by segmenting speech into broad categories and then performing SV. This work demonstrates the significance of vowel/ non-vowel like segmentation of speech and then performing SV. The SV studies using only vowel like regions (VLRs), VLRs and non-VLRs by explicit segmentation and then by implicit modeling are presented. It is proposed that, vowel/non-vowel like segmentation is beneficial, especially under degraded condition.
S R M Prasanna, Professor in the Department of Electronics and Electrical Engineering (EEE), IIT Guwahati, obtained his PhD from IIT Madras in 2004, MTech from NITK Surathkal in 1997 and BE from SSIT Tumkur (Bangalore University) in 1994. His area of interest is speech processing where he has a number of research contributions. He along with his colleagues and students has incubated a company SpeecHWareNet at IIT Guwahati that focuses on developing speech technologies for industrial needs.