Research On Speech Signal Processing Based On Auditory Model And Transductive Confidence Machine

Language is a tool that people use it to transfer information and daily communication.Using language makes human beings different from animals.In-depth studies of speech signal processing shows that human beings' auditory nervous system can't be reached by any modern information technology in either distinguish objects or determining orientation.This thesis presents an auditory bionic model based on biological structure of human ears.This model could extract and process speech signal to calculate characteristic value.Experimental results on transductive confidence machine(TCM)show that the recognition performance of our model is more accurate and sensitive.The primary work and achievements are as follows:Firstly,this thesis presents an auditory model of human ears.This thesis studies the biological structure of human auditory system and each part of human ears.The auditory model combines different auditory model,it includes the external/middle ear model,eardrum model,cochlear outer hair cells model and cochlear inner hair cells model.The external/middle ear model is mainly used for voice signal sampling,quantization and pre emphasis;Eardrum model is mainly completed frequency decomposition of the speech signal;The main function of cochlear outer hair cells model suppress noises and enhances speech signal;Cochlear hair cells model converts the mechanical energy of sound waves into biochemical energy,produce bio-energy,and transfer speech information to auditory nerve.Secondly,this thesis presents a classification model based on TCM.This model could give the identification results and its reliability,so the results are more convincing.Traditional TCM suffers from the limitation of less identification categories and low identification rate.This thesis presents an extended model of TCM to address this problem.The extended model transforms multi-class identification into two-class identification to simplify the identification problem.Finally,this thesis presents a speech signal processing model integrated auditory model with TCM.This model uses auditory model to extract parameters of speech signal and verify the characteristic value.The experimental results show that the integrated model has better performance on noise suppression and speech enhancement.Also,it could improve the identification rate and reliability.