| With the rapid development of scientific technology, new marine digital aids, such as ECDIS(Electronic Chart Display and Information System), AIS(Automatic Identification System) and VDR(Voyage Data Recorder), have being installed and applied on board at present. This implys that the ship's navigation at sea has gradually stepped into the epoch of "marine digital transport". VDR, as one of connotations of the future "marine digital transport", is a marine aid for recording the ship navigational data in real-time, which can be downloaded and replayed with special appliance or software. These data, especially bridge communication under emergency, are very importantly helpful and meaningful to analyze the marine accident and determine the responsibility when the ship accident occurred. It is well known that the sea state and situation are complicated and changeable when the ship proceeds at sea. When VDR continuously gathers bridge communication, playback is often vague and in the meanwhile the very large number data also will be stored in VDR. Hence, on the premise of the quality of playback data, how to effectively enhance and compress the bridge speech communication during the process of the data storage and transmission in order to increase speech understandability and save the storage volume, which are the most essential issue to be solved for VDR's development and research. During the rsearch, as per the successful application in speech signal enhancement and compression with wavelet analyse nowadays, this dissertation has maken new research and presentation to the speech signal enhancement and compression for VDR in detail.During the research of the bridge speech communication enhancement, this paper proposed to utilize the speech enhancement method of wavelet thresholding MMSE(Minimum Mean Square Error) estimation for spectral amplitude algorithm for VDR. This algorithm adopted a kind of thresholding function which has infinite order continuous derivative and overcomes the shortcoming of traditional thresholding function (hard-thresholding functionand soft-thresholding function). Then based on SURE(Stein's Unbiased Risk Estimation) unbiased estimation, we got a wavelet thresholding denoising method which the optimal threshold parameters can be searched adaptively. It can meet the requirement of time various signal in real-time processing. It is proved that this algorithm can effectively increase SNR and is much more effective and superior to tranditional ones by the experiment.One of the obvious differences between the wavelet transform and Fourier transform is that the wavelet function is not unique in wavelet transform. Different wavelet base produced by the different wavelet functions has different time-frequency characteristics, and are used to analyze the same issue will lead to the different result. Thereby, the choice of wavelet function in the application to research practical issue is one of the critical questions firstly. In this dissertation, Daubechies wavelet, the support orthogonality wavelet function with the most vanishing moment was selected according to the characteristics of different wavelet functions. After SNR increment and energe comeback degree in front and at the back of speech enhancement were tested, analysed, compared through the method of wavelet thresholding MMSE estimation for spectral amplitude, the final conclusion of wavelet function, db7, adopted mostly by this dissertation has been drawn.During the research of bridge speech communication compression, one self-adaptation wavelet packet transform method was ultilized to compress the VDR bridge speech communication after it was enhanced. The adapted wavelet packet transform based on R-D rule which selects the best wavelet packet base according to R-D rule, in which best quantization method and Huffman coding are used. This method produces the best compression version such that the error upon reconstruction is within an upper bound fixed a given value and obtains satisfactory compression results. |
PDF Full Text Request