Desenvolvimento de um sistema de reconhecimento pessoal de fala baseado em características prosódicas do sinal de voz.

Abstract

This work presents the development of a differentiated method for personal speech recognition, whose main purpose is the use of prosodic information of the signal, such as energy, textit pitch and autocorrelation to recognize phonemes. Its implementation is divided in creating a phonetic segmentation system and the phonetic recognition system itself. The segmenter is mainly used as input to the speech recognizer, and therefore its development is important because it is responsible for separating all phonemes found in a phrase. This system is not dependent upon the speaker and does not require training phase, unlike those already on the market. The speech recognizer proposed here is responsible for identifying which phrase, word or phoneme was pronounced by the speaker under test. Finally, the work explains results of both systems and compares them with current data, in order to be able to improve the programs developed with the addition of recent methods.