Analysis of Computer Visualized Sound Parameters of Vocal Music Singing Based on Deep Learning

Abstract

The first impression of the early historical recordings on the contemporary audience is often not ideal; the overall speed is relatively fast, the rhythm is relatively loose and free, and the strings are slippery. In recent years, a large number of research achievements have emerged at home and abroad that apply computer visualization analysis methods to music performance practice. It is feasible and necessary to gradually apply the visual audio parameter analysis method to music research and performance practice teaching. In this study, a prosody hierarchy prediction model based on the CNN (convective neural network) is proposed, and word vectors are added as semantic features. The DL (deep learning) method is applied to vocal music recognition, and a recognition method based on the DL framework is proposed by combining traditional audio signal processing methods. After introducing word vectors as features in the CNN model, the F-score value increased from 77% to 80%. The feasibility of the proposed vocal music recognition algorithm based on DL is verified by experiments.