Research On Key Technologies Of Speech Recognition Based On Shipboard Environment

With the advancement of human technology,speech recognition technology has been widely used in human-computer interaction.However,the current speech recognition technology still has various problems in the differentiated specific environment.Therefore,this article focuses on the research of speech recognition technology in the specific environment of the ship,explains the specific process and framework of speech recognition,and comprehensively describes the basic theories of core technologies such as speech signal preprocessing,endpoint detection and speech noise reduction,at the same time,certain improvements are proposed to some key links in the speech recognition technology:An improved short-time energy-zero-entropy feature parameter speech endpoint detection method is proposed.This method combines the short-term zero-crossing rate,short-term energy and spectral entropy of the characteristic parameters of the three speech signals and proposes a new characteristic parameter,which is the short-term zero-entropy value.Detect the endpoints of the voice signal by combining the new feature parameters and the dynamic dual-threshold method.This method has good endpoint detection performance in specific situations on ships.Propose a new algorithm for speech noise reduction using multi-taper spectrum estimation and improved Wiener filtering characteristics.The algorithm first estimates the power spectrum of the speech signal interfered by noise through the multi-taper spectrum method,thereby obtaining the prior signal-to-noise ratio at the corresponding frequency point,then use the Wiener filter method to reduce the noise of the speech signal.The algorithm can better eliminate a large amount of residual interference noise after Wiener filtering in the shipboard environment.Analyze and research the algorithm characteristics of Hidden Markov Model(HMM),and use MATLAB software to build a voice recognition system for shipboard commands.Show speech processing and speech recognition processes,apply the improved endpoint detection and speech noise reduction methods proposed in this article to compare experimental data with traditional methods.Through experimental verification,the improved new algorithm can improve the recognition accuracy of the speech recognition system in the shipboard environment,and has certain practical value.