Spoken language is used by people to communicate naturally with one another. A simplistic view of the communication process is as follows. Person A wishes to communicate an idea to person B. The idea, initiated in the mind/brain of person A is encoded into speech signals by means of the person A's speech production mechanism, the vocal apparata in the vocal tract. Various kinds of noise may interfere with the speech signals as they travel to person B. The resulting signal is captured by person B's speech receiving mechanism, the ear. It is then analysed and decoded into a meaningful message by the brain of person B. This thesis concerns itself with the investigation of and attempt to automate the receiving and decoding of English sentences using a machine - that is to perform the task of person B in the above scenario using a computer. The aim is not only to produce a sequence of phonetic sounds, but to look at the problems of building in the 'mind of the machine', a picture of the meanings, intentions, absurdities and realities of the spoken message. The various models, algorithms and techniques of speech recognition and speech understanding systems are examined. Speech signals are captured and digitised by hardware. The digital samples are analysed and the important distinguishing features of all speech sounds are identified. These are then used to classify speech sounds in subsequent spoken words. The way speech sounds are joined together to form syllables and words introduces difficult problems to the automatic recognition process. Speech sounds are blurred, overlapped or left out due to the effects of coarticulation. Finally, natural language processing issues, such as the importance of syntax (the structure) and semantics (the meaning) of sentences, are studied. A system to control and unite all the above processing is considered. The blackboard expert system model of the widely reported HEARSAY-II speech recognition system is reviewed as the system with the best potential for the above tasks. / Thesis (M.Sc.)-University of KwaZulu-Natal, 1992.

Computer speech recognition techniques were investigated. This investigation included a study of the hearing and speech process. An algorithm was developed that used nine features to identify the phonemes in speech signals. Two of these features, the total energy and the number of zero crossings in a specific section of the speech signal, were obtained directly from the digitized speech signal. The other features, frequency energy bands and formant frequencies, were measured from a spectral analysis of the signal. A Hewlett Packard mini-computer was used for the development of the necessary software in FORTRAN. For the testing of the algorithm ten words, "zero" through to "nine" were used. / Thesis (M.Sc.)-University of Natal, 1985.

Speech recognition systems fall into four categories, depending on whether they are speaker-dependent or independent of speaker population and on whether they are capable of recognizing continuous speech or only isolated words. A study was made of most methods used in speech recognition to date. Four speech recognition techniques for speaker-dependent isolated word applications were then implemented in software on an IBM PC with a minimum of interfacing hardware. These techniques made use of short-time energy and zero-crossing rates, autocorrelation coefficients, linear predictor coefficients and cepstral coefficients. A comparison of their relative performances was made using four test vocabularies that were 10, 30, 60 and 120 words in size. These consisted of 10 digits, 30 and 60 computer terms and lastly 120 airline reservation terms. The performance of any speech recognition system is affected by a number of parameters. The effects of frame length, pre-emphasis, window functions, dynamic time warping and the filter order were also studied experimentally. / Thesis (M.Sc.-Electronic Engineering)-University of Natal, 1990.

No description available.

No description available.

No description available.

Neural networks have recently been applied to real-world speech recognition problems with a great deal of success. This thesis developes a strategy for optimising a neural network known as the Radial Basis Function classifier (RBF), on a large spoken letter recognition problem designed by British Telecom Research Laboratories. The strategy developed can be viewed as a compromise between a fully adaptive approach involving prohibitively large amounts of computation, and a heuristic approach resulting in poor generalisation. A value for the optimal number of kernel functions is suggested, and methods for determining the positions of the centres and the values of the width parameters are provided. During the evolution of the optimisation strategy it was demonstrated that spatial organisation of the centres does not adversely affect the ability of the classifier to generalise. An RBF employing the optimisation strategy achieved a lower error rate than a multilayer perceptron and two traditional static pattern classifiers on the same problem. The error rate of the RBF was very close to the theoretical minimum error rate obtainable with an optimal Bayes classifier. In addition to error rate, the performance of the classifiers was assessed in terms of the computational requirements of training and classification, illustrating the significant trade-off between computational investment in training and level of generalisation achieved. The error rate of the RBF was compared with that of a well established method of dynamic classification to examine whether non-linear time normalisation of word patterns was advantageous to generalisation. It was demonstrated that the dynamic classifier was better suited to small-scale speech recognition problems, and the RBF to speaker-independent speech recognition problems. The dynamic classifier was then combined with a neural network algorithm, greatly reducing its computational requirement without significantly increasing its error rate. This system was then extended into a novel system for visual feedback therapy in which speech is visualised as a moving trajectory on a computer screen.

Thesis (MEng(TelecommunicationsbyResearch))--University of South Australia, 2003.

Thesis (M.S.)--Rochester Institute of Technology, 1989. / Includes bibliographical references (leaves 20-21).

Thesis (M.S.)--Ohio University, June, 2001. / Title from PDF t.p.