Research On Denoising Technique Of Noisy Speech Signal Based On Vehicle-mounted Digital Walkie-talkie

Digital walkie-talkie is the development trend of the walkie-talkie. It can provide users with more high-quality wireless voice communications.As the vehicles and vehicle-mounted terminals are getting more and more popular, the vehicle-mounted terminal that has an interphone can illustrate a good prospect of application. It must complete data collection, pre-processing , coding and other tasks of the speech signal at the sending end of the vehicle-mounted terminal during the speech communications.In the wireless speech communication, the clean speech signals are frequently disturbed by various environmental ambient noises and internal electrical noises during the process of acquisiton and transmission.These noises strongly influence the acoustic characteristics and the model parameters, then, the wireless speech voice playback have a serious impact.Therefore, it is great significant to study on noise reduction techniques of noisy speech signals for improving the quality of wireless communication.By the request of developing a vehicle-mounted terminal with interphone for a company, in this thesis, it firstly proposed an innovative concept of vehicle-mounted digital walkie-talkie with making use of the GPRS service of the 3S intergrated vehicle-mounted terminal to achieve instant speech communication.Then in allusion to the application environment , it puts forward the existing problems in realizing speech communication with good quality.Afterward combined with background, it introduces the history and the dynamic development of walkie-talkie and the techniques of noise reduction of noisy speech signals.Detailedly analysed the time-domain and frequency-domain properties of the speech signals and the noise signals according to the environment of the devices' application.By means of the speech generation model, we build a noisy speech generation model and a pretreatment model to make ready for getting accurate parameters of the sound model in environment with high levels of ambient noise. In this paper it has a in-depth study on the short-term analysis technique of the speech signals , pitch estimation methods and speech enhancement techniques.Considering hardware resouces are limmted, it is specific to do research on the pitch estimation methods based on short-term self-related function and programmed with C;discuss the subtraction spectrum and adaptive cancellation which have simple principles and small amount of operators in non-parameter methods.At the same time, it does research on the algorithm which used the cutting-edge wavelet transform to remove the noise. Aiming at problems existing in three algorithms, such algorithms are improved for better denoising effects.To verify every improved algorithm, the improved altorithms are simulated with MATLAB and comparative to the originals.Simulation results show that: the improved adaptive cancellation solves the problem of multi-channel to streamline the structure of the hardware, and the new LMS variable step algorithm can speed up the adjustment of the filter weights to increase SNR;improvement of the subtraction spechtrum can reduce the musical noise to increase SNR greatly ; multi-threshold spectrum subtraction in wavelet-domain is superior to the noise reduction with separated wavelet transform and spectrum subtraction in frequency-domain.Although the SNR of the voice getted through multi-threshold spectrum subtraction in wavelet-domain was smaller than that through spectrum subtraction, hearing the voice , the former is more comfortable and the MOS of the former is higher than the latter.The spectrogram of the speech signal with improved wavelet denoising is the most clear in all of three.