Desenvolvimento de um protótipo para reconhecimento de voz

Abstract

The effect of recognizing a person by his voice, through a machine, is known as automatic speaker recognition. This technique sets up a complex problem when considering algorithms that demand a fast processing and the different fields of knowledge, like signal processing, pattern recognition, acoustical physics, linguistics, mathematics, computer science, and others. This work has the purpose of investigating voice recognition systems, specificaly, automatic speaker recognition systems obtaining, as a research result, the development of a prototype. The prototype, based on speakers identification in an text independent classification, uses the Mel-Frequency Cepstrum Coefficients (MFCC) extraction to obtain the characteristic voice parameters and the LBG (Linde-Buzo-Grey) Vector Quantization algorithm as a technique for recognition based on pattern comparison. The whole system was implemented in ANSI C language running on a PC. The speaker recognition systems can be used in many types of practical applications such as banking, access control to restricted areas, access control in software, electronic signature, etc.