Speaker Segmentation Using Non Linear Energy Operator Based Variance Spectral Flux

Abstract

Abstract
Speaker segmentation is considered to be a process that attempts to find speaker segment boundaries in a given audio stream. It can be used in various applications of speaker diarization, speaker indexing, word count etc. Classification of speech and non-speech can be obtained by traditional method of variance spectral flux (VSF). In this paper, we investigate the new techniques to perform the speaker segmentation task for multiple speakers in one audio stream without any prior knowledge of the identities or the no. of speakers. It uses non-linear energy operator based VSF with delta BIC and Kullback-Leibler (KL2) distance matrices. It improves the results of speaker segmentations.

Keywords: Bayesian information criteria, Kullback-Leibler (KL2) distance metric, non-linear energy operators, variance spectral flux

Cite this Article
Sukhvinder Kaur, Sohal JS, Neha Sharma. Speaker Segmentation Using Non Linear Energy Operator Based Variance Spectral Flux. Current Trends in Signal Processing. 2017; 7(3): 1–6p.