| According to the World Health Organization(WHO)report,more than 360 million people in the world have disabling hearing loss.At the same time,because of the miuse of ears,the number of the hearing loss is still growing,and it will break through 900 million within the next 10 years.However,the traditional hearing aid is expensive and can't meet the demands of all users.Therefore,it is meaningful to develop an intelligent hearing aid system with a good user experience integrated with a variety of voice processing functions based on an appropriate price embedded platform.At the same time,from the perspective of changing the mode of human hearing,we design a new intelligent hearing system which combines speech modeling,recognition and separation technology.This system can change auditory state from passive to active,and everyone can use it to hear who he want to listen to,shielding all he didn't want to listen.This new idea can provides a new way of thinking about the future intelligent system.Therefore,in this thesis,taking speaker recognition,sound localization and Mixed voice separation as basic theoretical basis,the researches are focus on getting target voice intelligent-ly in complex voice environment.The main researching contents are listed out as follows,(1)For obtaining target voice information intelligently in complex voice environment,a system framework is designed,in which an intelligent hearing aid system is implemented by three modules:Speaker voice modeling;Voice sources localization in mixed voice environment;Mixed voice separation and extraction.(2)For speaker voice modeling,a research on the robust speaker modeling method is carried out.The reliability of the speaker voice model in low SNR environment is improved by adopting speach endpoint detection in voice pretreatment part.(3)For the extraction of target voice under complicated environment,the research on extracting the target voice according to the results of the identification of the voice source based on the theory of fanaticism,separation and speaker recognition is conducted.The mixed sound separated method whichis based on the sound intensity vector distribution is researched,and then the target sound signal can be extracted by recognizing the separated signals.(4)In order to determine the speech sources in the complex environment,a research on multi-sound-source localization and prediction is proposed,which is based on the decomposit-ion of the matrix feature space MUSIC(Multiple Signal Classification Multiple Signal Classification)algorithm.In this thesis,a speech source localization and real-time prediction and tracking method based on kalman filter and particle filter is proposed.The rationality and accuracy of the two methods is verified by experiments.Finally,a comparative analysis of these methods is provided,and the advantages and disadvantages of these two methods are given.(5)In view of the system hardware platform implementation issues,the intelligent hearing system is implemented on TMS320DM642 hardware platform.All the algorithms are completed from the PC platform to the embedded DSP platform,and the feasibility of intelligent hearing system can be validated.It can lay the ground work for the intelligent hearing aid system productization.In summary,a new idea about wearable devices is provided,which is from the perspective of changing auditory model,and aims to change people's hearing habits.This new method also provides new knowledge base for the development of speech processing technology. |
PDF Full Text Request