Research on Modeling of Vocal State Duration Based on Spectrogram Analysis

In the early stage of vocal music education, students generally do not understand structure human body, and have doubts about how to pronounce their voices scientifically. However, with continuous development computers, computer technology has become more developed, processing speed been greatly increased, which provides favorable conditions for application spectrum analysis in teaching. this paper, we first study GMM-SVM DBN, combine them extract deep Gaussian super vector DGS, further construct feature DGCS on basis DGS; then convolutional neural network (CNN), achieved great success image recognition task recent years, design a CNN model fusion features music. The experimental simulations show that fusion-based speaker system achieves very good results terms rate.