A Novel Coding Scheme for Keyword Spotting

Most of the works on keyword spotting have employed two HMM models, keyword model and nonkeyword model. However, it is apparent that it is quite inconvenient to form a model that covers all the out-of-vocabulary (OOV) words uttered during the practical applications. This paper proposes a different algorithm, which, without modeling the OOV words, exploits only the acoustic particularities of the predefined keywords to detect their frontiers in a sentence. As a result, this algorithm can achieve almost the same performance while consuming low computational and storage resources.