Research And Realization Of Sign Language Adaptation Based On MLLR

Sign language is the primary method by which hearing-impaired people communicate with each other.It conveys ideas by gesture and motions of both hands. Researches on the sign language recognition not only enable the communication between hearing-enabled and deaf people free of obstacles,but also enhance the computer's capacity of understanding body language.So far,the researches on sign language have lasted for more than 20 years.Although a lot of achievements have been made,there are still some challenging problems that need to be solved.For example,the Signer Independent Recognition is the bottleneck when the theory of sign language recognition is put into practice.Since the difficulties in obtaining enough gesture data from different signers and the apparent differences of the signers' gesture data,it's impossible to establish a common model set matching everyone well.To solve it,we learn the adaptation technology from the speech recognition field and apply it to the Sign Language problem.In this paper,a singer independent(SI) model according to the Hidden Markov Model is built for each sing word using the training data from several signers.And then we use a small amount of new signer's gesture data to modify the parameters of the SI model according to the MLLR algorithm.Experiments have been performed on both static and dynamic gestures set,and the result shows that applying adaptation technology to sign language recognition filed can improve the system's recognition precision to some extent.