Font Size: a A A

Research On Speech Recognition Based On Sentence Pattern Library

Posted on:2007-12-19Degree:MasterType:Thesis
Country:ChinaCandidate:C X ShiFull Text:PDF
GTID:2178360212457264Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Speech recognition technology is growing recognition currently. It is expected that the technology will come to pervade every aspect of our lives, such as industry, household electrical appliances, communication, automobile, medical treatment, household service, electronical production and so on. In its applications, speech recognition technology which based on special field and run on embedded system has become a hot technology now. Speech recognition technology based on PC needs enormous search space, but for the hardware limit in embedded system, the processor can not meet requirements of real time recognition. To solve the difficulty, improving on recognition algorithm is necessary.In this paper, prevailing speech recognition technologies are discussed, and the key to the technologies are introduced in detail. After that, the paper uses a method based on sentence pattern library, designs its thoughts, and implements this method by modifying the recognition algorithm of Julius&Julian, which is open source software in Japan.Speech recognition based on sentence pattern library is applied to one special field. Its main process is: analyses corpus of this field's language at first, extracts sentence patterns to make sentence pattern library, then transforms the library to determinate finite automaton (DFA) by the sentence patterns. During reconition, it combines of Viterbi-Beam algorithm, expands words'paths via the recognition space determined by DFA. So it greatly reduces the search paths' space.The paper makes experiments on a special task to get evaluation of recognition performance. After experiments by speech recognition based on sentence pattern library and Julius&Julian, the paper analyses experiment results, and draws a conclusion that speech recognition based on sentence pattern library can recognize faster and occupy less memory, at the same time it can keep high recognition rate. Because of that, in the application to embedded system, this method has certain advantage. Finally the paper makes a summary of speech recognition based on sentence pattern library, points out the current problem and future directions of improvement.
Keywords/Search Tags:Speech Recognition, Sentence Pattern Library, Determinate Finite Automaton
PDF Full Text Request
Related items