Multi-instrument signal Recognition on account of con-volutional NN

Abstract:
Aiming at the low accuracy of multi instrument recognition, a multi instrument signal identification system ConvNet on account of con-volutional NN is proposed. The basic principle and network structure of con-volutional NN are analyzed. Then IRMAS data set is used, It is divided into training and test data sets. To improve the accuracy and precision of the instrument signal recognition more, need to data preprocessing of the data collection, data processing after the success, ConvNet instrument recognition system is built, through this system implementation on the violin, piano, guitar and other instrument signal classification. The results indicate that the con-volution NN multi instrument signal recognition system proposed in this paper has high recognition accuracy. Under different analysis windows, it can achieve better recognition results, which shows that the system designed in this paper is effective.