Feature Mouth Shapes Extraction Based on Contour of Internal Lips

Feature mouth shapes extraction decide the recognition effects in speaker identification. For the problems of huge information and low recognition rate, a mixed method of feature extraction based on contour of internal lips was proposed. Firstly, feature points in contour of internal lips were selected to build basic mouth shapes. Secondly, general models of the training set were constructed by calculating mean values of coordinates. Then, Gabor wavelet transform was applied to extract feature information from basic mouth shapes and general models. Finally, according to similarity comparison between the general models and basic mouth shapes, the feature mouth shapes could be selected to identify the speakers. The experiments indicated that internal lips was more sensitive to deformation than outer lips and could have more feature information and higher recognition rate.