Catálogo de publicaciones - tesis
Título de Acceso Abierto
Optimización mediante algoritmos evolutivos de la representación de señales para el reconocimiento automático del habla
Leandro Daniel Vignolo Hugo Leonardo Rufiner Ignacio Ponzoni Omar Chiotti Pablo Granitto Diego Humberto Milone
acceptedVersion.
Resumen/Descripción – provisto por el repositorio digital
The key issue on speech recognition is given by the characteristics of the signals involved, as these are governed by complex probability density functions, are non-stationary and generally contaminated with noise of diverse nature and intensity. This is why the automatic recognition systems need a processing stage in order to bring out the key features of phonemes, allowing to improve their performance. The goal of this thesis is the development of a methodology for the optimization of the signal processing stage, in order to improve the results of an automatic speech recognition system. This methodology consists in the use of evolutionary algorithms for the optimization of the feature vector used for speech signal representation. The hypothesis is that the better the analysis or process applied to the patterns that are to be classified, the more separated would the classes result in the features space and, therefore, the classification task would be simpler. In this thesis, the first proposal is to continue the search for an optimal representation based on cepstral coefficients, by the optimization of the filterbank involved in this feature extraction procedure. On the other hand, wavelets have characteristics that are useful for the analysis of non-stationary signals. These features present discriminative information, however, the large number of coefficients makes the task of the classifier more difficult. Because of this, the use of an evolutionary algorithm is proposed to search for a subset of coefficients which maximizes the discrimination capability.Palabras clave – provistas por el repositorio digital
Algoritmos evolutivos; Cuantización vectorial; Modelos ocultos de Markov; Paquete de onditas; Coefb01cientes cepstrales; Reconocimiento robusto del habla
Disponibilidad
| Institución detectada | Año de publicación | Navegá | Descargá | Solicitá |
|---|---|---|---|---|
| No requiere | 2012 | Biblioteca Virtual de la Universidad Nacional del Litoral (SNRD) |
|
Información
Tipo de recurso:
tesis
Idiomas de la publicación
- español castellano
País de edición
Argentina
Fecha de publicación
2012-02-28
