Treinamento multi-estilo e adapta??o de modelos via MAP para reconhecimento de fala em ambientes ruidosos

• Ano de defesa: 2011
• Autor(a) principal: Val?rio, Thales Antonio Fernandes
• Orientador(a): Ynoguti, Carlos Alberto
• Banca de defesa: Klautau Junior, Aldebaro Barreto da Rocha , Ynoguti, Carlos Alberto , Brito, Jos? Marcos Camara
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Instituto Nacional de Telecomunica??es
• Programa de Pós-Graduação: Mestrado em Engenharia de Telecomunica??es
• Departamento: Instituto Nacional de Telecomunica??es
• País: Brasil
• Palavras-chave em Português: Reconhecimento de fala; modelos ocultos de Markov; compensa??o de modelo baseada em ru?do, adapta??o; treinamento adaptativo
• Área do conhecimento CNPq: Engenharia - Telecomunica??es
• Link de acesso: http://tede.inatel.br:8080/tede/handle/tede/111

The accuracy of voice recognition systems degrades severely when operating in noisy environments. One of the causes pointed out in the literature for this fact is the acoustic mismatch between the environment in which the training locutions were recorded and the one in which the speech recognition system operates. From the mathematical modeling of the inuence of the environment on the speech signal, two ways of improving the rate of correctness of an automatic speech recognition system in these environments were evaluated: multi-style training, using material corrupted by noise Varied but of low intensity to train the system and adaptation of the acoustic models through the maximum a posteriori method. This combination raised the system hit rate by an average of 23.67%.