Keyword spotting(KWS) is an important area of speech recognition. Its objective is to spot the given keywords from continuous speech. Comparing with the continuous speech recognition, KWS has advantages of less resource consuming, higher accuracy and stronger practicability. This technology has practical application in many areas such as surveilance, content-based retrieval and telephone response.There are two steps in a KWS system. Putative hit is recognized, and then it is verified. This paper mainly focuses on a method of utterance verification which uses immune algorithm.Firstly, this paper generally introduces the theory and framework of KWS. Secondly, the mechanism and algorithms of artificial immune system(AIS) are researched and a new immune algorithm is proposed. Using this algorithm, an AIS-based classifier is established for each keyword. A putative hit is judged by the post-classifier of its corresponding keyword. A hybrid HMM/AIS based system for keyword spotting is built in this paper, and the multi-decoding mechanism is realized.Experiment results show that HMM/AIS-based keyword spotting system achieves a good performance, which provides a new solution to utterance verification. |