Research On Speech Recognition Based On VQ&Fulltext Retrieval

With the continuous development of society, governments pay more and more attention to protect the interests of the population occupies a significant proportion of disability groups, "equal access to information" as the basic human rights received more recognition, and information accessibility and its process of internationalization has become a symbol of social civilization and progress.In China, the government has realized the significance of information and the importance of accessibility , mealwhile carrying out research of barrier-free information which relevant policies, regulations and standards of . Some of the information industry leader such as IBM have also take steps in heir products and services to develop and implement the concept of information accessibility. During rapid development of network information age, the handicapped need more extensive information knowledge, Most people regards networks as an important way to study information, so accessibility information has become vital significance. However, physical and mental defects cause that the handicapped have difficult in collecting information and learning knowledge. It must impact on their study, employment, social stability and harmony. So how to make the information collection and query service implementation accessibility is particularly important.This paper presents three-tier voice search prototype system based on vector quantization (Vector Quantization, VQ) and Lucene , objective of the paper combining two technologies to design and develop a auxiliary information collecting, querying prototype system which has open source, modular, scalable attributes. To enable disabled persons can collecte and inspect information through the system, finally accomplishing the goal information accessibility for disabled.Mainly works in this paper :i. First of all ,The current status of the voice search is introduced. mainly summarize speech recognition technology, search engine technology, analyze the key technology and illuminate the principle of implementation;ii. Followed, This paper using dynamic, random method to select the parameter in a certain range, and combined with differential cepstrum parameters to the establish speaker recognition model。The results show that the method will improve the recognition rate by the circumstances,the training time and the recognition time has improved significantly;iii. Finally On this basis, complete the analysis and the design of prototype three-tier system based on VQ speech recognition technology combine with the Lucene full-text indexing / retrieval technology, summarize the system design program, divided the system into modules and finish each module detailed design and implementation.iv. During the phase of design and analyze Lucene module, through the experimental test time performance of setting up index, mealwhile analysis index and its optimization.