SVM decision taking in I-Vector based Speaker Verification Systems
From LRDE
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.
- Authors
- Benjamin Roux
- Type
- techreport
- Year
- 2013
- Number
- 1315
Abstract
State of the art speaker verification systems provide a decision based on cosine distance. A different approach is to consider a decision based on Support Vector Machine (SVM) separation. SVMs have been largely used in pattern recognition and decision making as well as in combination with GMM super-vectors (GSV). However, the idea of using SVMs in the i-vector space for speaker recognition has not been well explored. We will study the different kernel functions and how it could be used in speaker verification with i-vector space. We will explore how we could deal with channel and speaker variabilities in the feature space. The experiments are done on NIST-SRE 2010 core condition database.