Fractal based speech recognition and synthesis
Date
Authors
Advisors
Journal Title
Journal ISSN
ISSN
DOI
Volume Title
Publisher
Type
Peer reviewed
Abstract
Transmitting a linguistic message is most often the primary purpose of speech communication and the recognition of this message by machine that would be most useful. This research consists of two major parts. The first part presents a novel and promising approach for estimating the degree of recognition of speech phonemes and makes use of a new set of features based fractals. The main methods of computing the fractal dimension of speech signals are reviewed and a new speaker-independent speech recognition system developed at De Montfort University is described in detail. Finally, a Least Square Method as well as a novel Neural Network algorithm is employed to derive the recognition performance of the speech data. The second part of this work studies the synthesis of speech words, which is based mainly on the fractal dimension to create natural sounding speech. The work shows that by careful use of the fractal dimension together with the phase of the speech signal to ensure consistent intonation contours, natural-sounding speech synthesis is achievable with word level speech. In order to extend the flexibility of this framework, we focused on the filtering and the compression of the phase to maintain and produce natural sounding speech. A ‘naturalness level’ is achieved as a result of the fractal characteristic used in the synthesis process. Finally, a novel speech synthesis system based on fractals developed at De Montfort University is discussed. Throughout our research simulation experiments were performed on continuous speech data available from the Texas Instrument Massachusetts institute of technology ( TIMIT) database, which is designed to provide the speech research community with a standarised corpus for the acquisition of acoustic-phonetic knowledge and for the development and evaluation of automatic speech recognition system.