An Improved Approach to Open Set Text-Independent Speaker Identification (OSTI-SI)

摘要

This paper focuses on open set text independent speaker identification which is one of the most challenging subclass of Speaker recognition. The initial stage is similar to closed set speaker identification, where the distortion for each test voice against all train voices are determined. The distortions after normalization is set as decision criteria which eases the process of thresholding. The threshold variation which is mostly independent of dataset but dependent on the size of train data set and its values are quite similar for three datasets. The identification rate with balanced False Acceptance Rate (FAR) and False Rejection Rate (FRR) is 73-86%.