Studying On Chinese Digital Speech Recognition Technology Based On Neural Network

Speech Recognition is a complicate non-linear process. And recognition method based on linear system theory, such as Hidden Markov Model, comes to show certain limitations. With the deep research on non-linear theory of Artificial Neural Network, speech recognition method based on neural network has become a popular research topic. In this paper, MFCC and LPCC with mixed parameters based on Neural Network are applied to research on speech recognition.Digital speech recognition based on Neural Network has been proposed in this work. Computational validation, performance analysis and results assessment are presented for preprocessing, feature extraction and recognition procedures as well. Several recognition approaches are compared. Design principles of Neural Network for speech recognition are discussed. And different parameter features' affects on performance are analyzed as well. Given the constructed recognition model and algorithms, an experimental software platform is designed and developed. Through simulation and computation, related algorithms are compared with each other, and the affects on recognition results such as different feature parameters, training samples, background noise and specific person, are analyzed in detail. Experimental results demonstrate that speech recognition methods proposed combining MFCC and LPCC with mixed parameters, has better recognition performance when comparing with single parameter methods of MFCC or LPCC. In reality, Neural Network has a good application prospect for its high recognition rate.