Analysis and Recognition of Cello Timbre Based on Deep Trust Network Model
Peng Sun · Journal of Physics Conference Series · 2020
Abstract Voice color analysis and similarity calculation of music signals are the important research contents of computer music information retrieval system. In this paper, the deep trust network model is applied to the study of musical tone model. The 72-dimensional features of the cello tone are first extracted. Using the wrapper feature selection method, a 14-dimensional optimal feature subset that reflects the tone characteristics is selected, which greatly reduces the complexity of cello tone similarity calculation. In the set, SVR is used to classify and distinguish eight types of tone data, and a recognition accuracy of 62% was achieved, which is verified the feasibility of the tone model.