Chinese Traditional Musical Instrument Computer Intelligent Recognition Based on Recurrent Neural Network

Abstract

Musical instrument recognition is an important task in the field of music information retrieval (MIR). It helps people search music and identify music types through musical instruments, making music retrieval more accurate. At present, most of the existing research on instrument recognition is oriented to Western instruments, while there is little research on Chinese traditional instrument recognition. To solve this problem, we propose a musical instrument recognition model RMIR-Net based on recurrent neural network (RNN). For feature extraction, we combine Mel frequency cepstral coefficients (MFCC) with its difference coefficients to take full advantage of the static and dynamic characteristics of the audio. We create a diverse dataset of Chinese traditional musical instruments to verify our model. Through experiments, we select leaky rectified linear unit (LReLU) as the activation function for our model. The experimental results show that our RMIR-Net has better accuracy than classical machine learning methods in Chinese traditional musical instrument recognition.