The Research Of Speech Keyword Retrieval Based On Uyghur

With the development of the People’s demand for the retrieval ofinformation,text retrieval has been far from satisfying the needs of users. They requiremore and more retrieval of audio, video and other files. The Uyghur voice keywordsearch research currently has an extremely important significance. In this thesis, themain job of the work of the following aspects:(1)Uyghur speech recognition as a voice of the Uyghur Keywords retrievepre-processing stage, It establishes the acoustic model and language model tocomplete the identification phase of work. During the procedure of selection onacoustic parameters, with the impact of noise the acoustic parameters must benormalization. Thereby improving the speech recognition rate.(2)Prosodic features of the language has played a pivotal role in thecommunication. So we study Uyghur prosodic features-accented, to extract theparameters related to the stress parameters using the feature fusion method, thisprocess increases the dimension of PCA,who needs dimensionality reduction andde-redundant.We may find a larger characteristic parameters to stressed affectionthrough the experiments’ comparison and analysis to improve stress detection rate.(3)In response to the existing problems in the Lattice,We improve ConfusionNetwork from lattice using a clustering method,which is similar to a linearstructure,reducing the top of the grid points and the number of arcs and improveretrieval performance.(4)In order to improve the retrieval speed, The thesis uses the technology of theinverted index.(5)Through various experimental aspects, We integrate Uyghur speech retrievalsystem using VS2010tools...