Robot Acoustic Parameter Extraction And Noisy Speech Recognition Based On Autoencoder

Speech recognition is one of the important ways of human-computer interaction.In recent decades,speech recognition technology has made remarkable progress,and it is also used in our daily life more and more.With the wide application of speech recognition technology,the demand for accuracy of speech recognition is increasing day by day.Deep learning is one of the most popular subjects in recent years with the development of computing technology.It can discover more essential features of data by simulating the structure of human brain.If deep learning knowledge is applied to the field of speech recognition,it will bring unexpected gains.Firstly,this paper introduces the overall framework and key steps of the speech recognition system,and then compares their advantages and disadvantages with each other,the simulation experiment is carried out to verify.This paper focuses on the speech recognition feature extraction,model training and recognition parts,the speech signal's LPCC parameters and MFCC parameters were compared separately.The performance of MFCC is better than LPCC,and then,compared the ability of anti-noise of different components,the result is that a characteristic parameter which mixed the static and dynamic feature can more effectively resist noise interference.At the same time,the paper focused on the application of denoising Autoencoder in speech recognition,including the enhancement in front-end features and speech recognition in back-end.The experimental results show that the deep denoising Autoencoder has good performance in improving the noise immunity of speech recognition.