As one special field in speech recognition research, keyword spotting is to determine occurrences of one or more keywords embedded in unconstrained extraneous speech and/or noise. It has bright future in many application areas and is an important direction. Many scholars have a study for it and bring in product.In this paper, we give a brief history keyword spotting research and provide a discussion of its fundamental principle in which two problems in this field are pointed out, that is speech signal feature and model. For speech recognition system, model is a important problem, as an feature vector reflexion a word in the lexicon, demarcate the different recognition unit with a good method. This paper describes the basic theory of HMM and presents simple and practical methods for building HMM, and build an one-phase detection-verification system. An alignment score about input speech and reference model can be obtained in the searching. That can detect keywords from continuous speech without grammar restriction.For a keyword spotting system, more based a hidden markov model, but the score and spotting method are different, building a grammar network including keyword model and garbage model representing non-keyword and background in common. A garbage model based syllable lattice is expounded in the paper for small wordage keyword spotting system, by low detection time and high diction rate. |