Research And Implementation Of Speech Recognition Technology Based On Neural Network

The speech recognition technology has had a profound impact on human’s way oflife since it comes out and thus has always been the focus of scholars from all over theworld. At present, the dynamic time warping technology and Hidden Markov model,based on the principle of probability and statistics are widely used in speech recognition,while using artificial neural network in speech recognition is a relatively new researchmethods proposed in recent years. Speech signal is a complex nonlinear process.Therefore, artificial neural network based on the nonlinear theory which hasself-adaptability, parallelism, robustness, fault tolerance and learning characteristics, isbecoming a new research direction. In this thesis, the author combine the BP networkwhich is most used in the neural network with speech recognition. The major tasks andachievements are as follows:Firstly, the thesis analyses the basic principles of speech recognition from its levelmodel and system model and gives the voice’s entire preprocessing process includ-ing the original signal’s acquisition, pre-emphasis, framing, adding window and end-point detection, and discuss the different methods of acquiring speech feature values,especially analyses the extraction of Mel Frequency Cepstral Coefficients.Secondly, the thesis introduces the basic structure and characteristics of the neuralnetwork and importantly analyses the three-layer feed-forward error back propagationnetwork, and gives the derivation of its standard algorithm, then analyses the defectsand deficiencies of the algorithm, and gives the improvements by adjusting the neuron’stransfer function on the basis of previous research. The specific method that has beenderived afterwards, is adding a temperature coefficient and a location coefficient to theactivation function which would make the network parameter has more information andfaster convergence rate. Momentum factor and batch mode of training are also adoptedin the improved BP algorithm, which has been proved to be effective by a simple func-tion approximation experiment.Finally, the author design a speech recognition simulation system based on BPneural network, using MATLAB and visual studio, and then complete the training and recognition by own voice. The system uses a time warping algorithm, which is used tocompress and combine the feature values for the need of backend BP neural networkwhich asks for the input data having the same dimensions. Through experiments, wecan get the following conclusions: the improvement learning algorithm is superior totraditional BP training algorithm on the recognition rate and speed of convergence; theMel Frequency Cepstral Coefficients based on the model of the human auditory hasbetter effects than the Linear Prediction Cepstral Coefficients based on the model ofspeech; the numbers of BP network’s hidden layer neurons have a large impact on thesystem recognition rate and need to experiment to determine the best values.