Study Of Articulators' Contribution And Compensation During Speech By Articulatory Speech Recognition

In this study,the contributions of dynamic articulatory information and the compensation between articulators are evaluated by using articulatory speech recognition system.The Electromagnetic Articulographic System observed articulatory data that is a relative small dataset and hard to be recorded comparing with popular speech corpus used for modern speech recognition study.We used articulatory data as inputs,including a number of specific articulatory channels,to study each observation point on the vocal tract's contribution in speech recognition by DNN framework.We also analyzed each kind of phoneme's recognition results according to their speech production rules.The contribution rate of each articulator can be considered as the crucial level of each phoneme in speech production.Further,the results indicate that the contribution of each observation point is not relevant to method.We also studied the compensation between the articulators by DNN-based articulatory Recognition.The articulatory organs consist of the lower jaw and soft-tissue organs of the tongue lips and velum.These are soft tissues that use muscle contraction.Tongues and lips have attachments to the lower jaw,so when we removed one channel's information,the other points compensated.The result shows that the tendency of each sensor's contribution is identical to the rules of Japanese phonology.There are differences in the articulatory of behavior for phonemes,which is the common feature among speakers.The tongue is more important in speech production and provides more information than other organs.Individual difference is also observed for the contribution of the tongue.It was discovered that the removed channel decreased the recognition rate of the phonemes whose highest crucial point is this removed channel.However,some phonemes' recognition results were not affected when removing some observation points.These phenomena can be explained that other sensors on the vocal tract compensated the lost information.So we can explore the crucial point of vocal tracts in speech production to help analyze the compensation for articulators,which will contribute to the study of speech production and articulatory-based speech recognition.