Specific two words lexical semantic recognition based on the wavelet transform of narrowband spectrogram

Abstract

This paper presents a method based on wavelet transform of the narrowband spectrogram for specific two words Chinese lexical recognition. In the process of image feature extraction, the image processing technique is applied to the speech recognition field. Firstly, two-dimensional discrete db4 wavelet is used to decompose the narrowband spectrogram, which is divided into 6 layers of wavelet package decomposition, and calculates the approximate energy value. Then, the extracted approximate energy value is divided into level detail energy value, vertical detail energy and diagonal detail energy value, sets respectively as the narrowband spectrogram of the first characteristic set, the second and third feature set. The above three feature sets are used as feature vectors to support vector machine as a classifier for the overall recognition of two words Chinese vocabulary. 1000 voice samples are used in the simulation experiment. The results show that this method correct recognition rate can reach 96 percent.