Text independent speaker recognition using hybrid ensemble and adaptive boosting techniques

Abstract

The most natural type of individual communication is speech. Speaker newlineRecognition (SR), a method that identifies speaking humans and distinguishes newlinethem using machines, is based on speech characteristics and acoustics. It has newlineapplications in disciplines ranging from human computer interaction, biostatistics, newlinesecurity and the Internet of Things. Voice signal variations make speaker newlinerecognition difficult. Speech variability is governed by a variety of parameters, newlineincluding the speaker s emotional state, age and gender, among others. The goal newlineof SR is to implement a framework that can convert an acoustic input into a newlinesequence of speaker class tags. The two primary components of an SR system are newlinea front-end processor and a back-end classifier. The front-end processor is newlineaccountable for obtaining hand-crafted speaker characteristics or attributes, which newlineare analyzed by the back-end classifier for speaker recognition. newlineWhen compared to some standard biometric schemes, voice is one newlinemetric that, in addition to being common to users, delivers equivalent and newlinesometimes even greater levels of security. Despite the fact that modern newlinedevelopments in Machine Learning (ML) and Deep Learning (DL) techniques in newlinespeech and speaker identification approaches have vastly enhanced performance, newlinehowever limited by their need for phonetic or/and spokesman tags in newlinethe background data. Tagged background records are difficult to come by in newlinepractice, particularly when huge amounts of training data are needed, with newlinenumerous ways to increase their identification rate and end-user utility. newline newline