Sistema para identificación de hablantes robusto a cambios en la voz

Abstract

The speech recognition systems consist of three principal parts: preprocessing, features extraction and vectors classification. This paper considers the voice changes, voluntary and involuntary changes, and how this affects the speaker recognition. In this project is explained the preprocessing of the signal and how are obtained the voiced segments. Also is applied a features vector based in voice properties and Mel Frequency Cepstrum Coefficients (MFCC), and a Support Vector Machine (SVM) and an Artificial Neural Network as classifiers. The experiments consist in analyzing the frame presented to the system, detect the voiced segment and indicate to the system which vowel has been said, for an identification of which person pronounce the vowel. The results show the features-MFCC vectors have high rate on recognition.