Research For Automatic Speech Recognition Basedon Wavelet And ANN

With the development of modern science and computer technology, people will confront difficulty with communication with more and more machines, so it is very useful to adopt language in intercommunion which apply speech recognition. Wavelet and Artificial Neural Networks(ANN) become main current technology in speech recognition.At present, ANN theory research daily tend towards thoroughly,which important development aspects is pay attention to integrate with Wavelet transform. Wavelet transform has the exceptional of temporal-frequency localization,whereas ANN have execellent characteristics of self-learning.falut-tolerance.self-adapt and generalization ability,so how to combine their good merits .which is the important problem for people to attention .thereinto a sort of means is to prceed pretreatmently with wavelet theory.then would pick-up eigenvector give to neural networks.In the text,we mostly carry through studying the application of the two technique in ASR.At first, wo discuss what to implement a integrity system of automatic speech recognition principle,and investigate key technique of speech recognition system,especiaUy for the three parts of speech signal catching, speech signal pretreatment and speech signal character picking up in recognition syetem, we carry through simulation and analysis.Secondly, combing with wavelet theory, we analyse the import process of wavelet in speech recognition pretreatment of wave part.Thirdly, the application of Artificial Neural Networks (ANN) to Automatic Speech Recognition(ASR) is investigated in this thesis. The recognition of speaker-independent and isolated words is focused and three types of ANN model are presented. The related algorithms and programs are developed. The characteristics of numerical and application for the methods are illustrated by using simulation testing.Especially, the influence of some factors,such as feature parameter, number of training samples.background noise and speaker independent or dependent, is discussed. Results show that ANN has a higher recognition rate and potential advantages in automatic speech recognition.In the end, we put forward a suit of integrity speech recognition system in using stable point DSP on PC flat roof, aiming at hardware implementing of speech recognition, emphatically dicuss the stable point DSP transplant of speech recognition algorithm, and simulatedly implement on PC.