Avaliação de diferentes tecnicas para reconhecimento da fala

AUTOR(ES)
DATA DE PUBLICAÇÃO

1997

RESUMO

This work presents an evaluation of speaker independent isolated word recognizers using Hidden Markov Models (Discrete, Continuous and Semicontinuous), Artificial Neural Networks (Multilayer Perceptron) and Hybrid Systems. All the recognizers were evaluated considering the same database. The goal of these comparisons is to identify the advantages and disadvantages of each technique used in speech recognition, considering the following features: training and recognition time, recognition accuracy, complexity of algorithms and others. It is also reported the result of a comparison among different algorithms used in word endpoints detection. Moreover, several distance measures employed in vector quantization were evaluated with regard to recognition performance. In addition, different kinds of parameters used to represent the speech signal such as LPC coefficients, Mel Frequency Cepstrum coefficients, PLP coefficients were considered in the evaluation of recognizers and it was discussed the efects of cepstral mean subtraction in order to improve the recognition accuracy. The best recognizer performance of 99.47% was obtained combining different features

ASSUNTO(S)

reconhecimento de padrões processos de redes neurais (computação) reconhecimento de palavras markov reconhecimento automatico da voz

ACESSO AO ARTIGO

Documentos Relacionados