Time-sequential Speaker Recognition System

Syroka Zenon,Cezary Łabarzewski,Tomasz Zając,Paweł Dubiłowicz

doi:10.13189/asp.2014.020401

Abstract

The method presented in this publication was submitted in a patent application in May 2011, the patent was granted on December 17, 2014. It is a continuation of speech signal recognition study [1] carried out at the UWM. In 2010 a "Speech analysis and recognition method and system" [2] has been developed using sequential analysis basing on the Walsh basis functions. The present system includes time-variability of the speech signal spectrum. Basic block diagrams describing the system's mechanics have been presented.

Highlights

The present paper shows a speaker recognition system based on application of join time-sequential speech signal analysis
The time-sequential speaker recognition system has been portrayed in fig.[1]
The signal along with ambient noise registered by the recorder (1) is subjected to time-sequential analysis in the analyzer (7), as a result speech signal's and ambient noise's time-sequential transform coefficients are calculated

Summary

Introduction

The present paper shows a speaker recognition system based on application of join time-sequential speech signal analysis. In speech signals' time-frequency analysis the basic functions used to “build” transforms are trigonometric functions. These functions are periodic and zero-crossings occur in equal time intervals. In this case, the term sequency is equivalent to the term frequency. In patent application P.390450 [2] submitted on 02.19.2010, titled “Speech analysis and recognition method and system” a method of speech recognition using Walsh's sequential function basis has been shown. It didn't take speech signal's spectrum's time-variance into account. Considering the spectrum's time-variance the system becomes more complex and the analysis time becomes longer, yet consideration of joint time-sequential features yields more precise speaker recognition results

Structure and Mechanics of Time-sequential Speaker Recognition System

Conclusions