Automatic Speech Recognition Based On AR-GARCH Model

Modern society has stepped into the era of"big data".The rapid development of computer technology has promoted the upsurge of artificial intelligence.Automatic speech recognition(ASR)is one of the applications of artificial intelligence,which is closely related to people's life.The existing automatic speech recognition technology is mainly based on deep neural network.However,the LPC parameters(AR Model)are used to describe the characteristics of the acoustic model.Although in the current mature speech recognition industry,AR model can also play a good role,but for the emerging dialect recognition technology,there is not enough large sample database,so we can consider optimizing the model algorithm to improve the recognition accuracy.This paper analyzes the previous improvements and thinks that the LPC parameters can be improved to AR-GARCH model,which can play an optimization effect.And BP neural network algorithm is introduced to test whether the improved accuracy is significantly improved.Firstly,90 vowels O and 90 vowels I from open source are selected as sample data,which are cut and denoised to improve the quality of sample data.Then,the AR(12)model,AR(12)-GARCH(1,1)model and ARMA(1,4)model are fitted to the samples respectively,and the parameter estimates of each sample are obtained,and the output is speech feature parameter sequence.The data sample set is composed of the parameter sequence and the discretized sample tag,which is divided into training set and testing set.The BP neural network under Tensorflow framework is used for training.The test results show that the ar-reach model has the highest speech recognition accuracy of 74.07%,61.11%,and 68.52%,respectively.It is proved that the recognition effect of AR-GARCH model is better than that of AR model and ARMA model.In order to further verify,this paper increased the number of samples,and finally found that with the increase of the number of samples,the recognition accuracy of the three models has a rising trend,and the accuracy is more and more close.It can be concluded that when the sample size is small,the effect of AR-GARCH model improving the accuracy is better.In general,AR-GARCH model is more stable andaccurate.