The Mandarin Continuous Speech Keyword Spotting System Medium Vocabulary

Speech recognition is such a kind of technology that deals with speech signals with computer and recognizes them to symbol sequence. The technologies of small or medium vocabulary speech recognition, such as voice command recognition, keyword spotting and continuous digit string speech recognition, are of great importance in the application-oriented study of speech recognition.In this dissertation we focus our research on Mandarin keyword spotting system based on Biomimetic Pattern Recognition theory. This theory emphasize "matter cognition" instead of "matter classification". The application in this dissertation is realized using software simulation while most other applications is represented using neural networks hardware. And a hyper-polyhedron covering algorithm is proposed. The experiment proves that keyword spotting system using Biomimetic Pattern Recognition can gain higher detection rate than the traditional Hide Markov Model methods with a small training set.And the application to the endpoint detection field also gets good performace.Keyword spotting system must have a utterance verification phase. In this dissertation the phase is realized by taking a classifier as verifier. After investigating the rigorous Support Vector Machine algorithm in Statistics Learning Theory, we proposed a new parameters selection algorithm. Relevent experiments prove that the classifier in utterance verification phase can improve the detection rate of system.By the synthesis of above research results, this dissertation realizes a keyword spotting system for travel information navigation.