An Investigation of Type-1 Adaptive Neural Fuzzy Inference System for Speech Reconigtion

• Autor(a) principal: Lima, Thales A.
• Data de Publicação: 2018
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFRN
• Texto Completo: https://repositorio.ufrn.br/handle/123456789/34182
• topic: Computação; Computing; ANFIS; ANFIS; Reconhecimento de voz; Speech Recognition; Fonemas; Phoneme; Conjuntos difusos; Fuzzy sets; Redes Neurais; Neural Networks

Using voice for user recognition is something that humans do since the beginning and it a very natural ability. Being able to recognise the user by its voice is very important, but, in some cases, being able to recognise what is being said automatically can have very interesting and useful security applications. Thus, speech recognition has been experiencing a increasingly growth in attention in the last years, following the advancements of the machine learning field. Since this is a very complex problem and can have interference from several different sources, there has been widely different approaches to perform this task, very often with high cost, and more frequent than not, with results that are dependent on the high quality of the data, which is not always the case. In this paper we present an Type-1 Adaptive Neural Fuzzy Inference System for speech recognition on MOCHA-TIMIT repository. Besides, we also used the Mel-Frequency Cepstrum Cofficient and Filter-Banks feature extraction methods aiming to translate speech to text with low or medium quality samples and still have a good results when dealing with speech recognition.