Position-specific posterior lattices-based open-vocabulary spoken document retrieval
Guo Hong-y · Journal of Yanshan University · 2013
A novel approach to open-vocabulary spoken document retrieval is proposed using word fragment-based position-specific posterior lattices.Query extension is introduced to generate multiple query candidates to compensate the error of ASR and user input error.During the retrieval procedure,the edit distance weight is used to adjust the relevance ranking computation via fuzzy matching.The experiment shows that the proposed method is better than traditional methods not only in out-of-vocabulary processing but in robust performance as well.