Research And Application Of Continuous Digital Speech Recognition System

With the continuous development of information technology,especially the great advance made in algorithms,cheap parallel computing,big data and other technology,artificial intelligence is leading the trend of IT development in the 21 st century.Intelligent human-computer interaction interface is a research hotspot of artificial intelligence.Many experts point out that,the dimension of human-computer interaction will continue to transit from "touch" to "voice" after the transition from "click" era to "touch" era.If man-machine interaction is realized,undoubtedly,the machine understands the human language and execute the human's intention according to the information will be an ideal man-machine interaction way,and the realization of all of this should be based on the research of speech recognition technology.In this paper,we focus on the study of non-specific continuous digital speech recognition.The purpose of this paper is to realize the automatic voice dialing function of IP telephony application.The main work and achievements of this paper are as follows:1.Had a better understanding of Mixtures of Gaussians(GMM),Hidden Markov Model(HMM)and its related algorithms.2.This paper studies the signal processing and feature extraction in speech recognition,including framing,adding-windows,pre-processing,VAD,feature extraction,and proposing a GMM-based VAD method.3.Combined with the recognition task,the existing data resources,and the computing power of PC and Android mobile phone,the Hidden Markov Model based on mixed Gaussian model(GMM-HMM)is selected to carry out acoustic modeling and design the topology of the model.4.Deeply appreciated and referenced to the open source code of the speech recognition system HTK of Cambridge University.By adjusting the complexity of Gaussian mixture model and optimizing the decoding network,this paper built a system which can recognize a string ofcontinuous numbers.Besides,this paper gave an in-depth study of continuous speech recognition principle.5.Transplanted the HTK recognition part and the trained model to the actual Android platform applications,and achieved voice dialing capabilities via an IP phone application,which made the recognition technology of this paper get practical application.