| 引用本文： |
-
龚谊承,刘青,肖浩逸.基于RF-BiLSTM-Attention音乐分类方法的京剧二分类仿真[J].广西科学院学报,2023,39(3):322-330,339. [点击复制]
- GONG Yicheng,LIU Qing,XIAO Haoyi.Beijing Opera Binary Classification Simulation Based on RF-BiLSTM-Attention Music Classification Method[J].Journal of Guangxi Academy of Sciences,2023,39(3):322-330,339. [点击复制]
|
|
| 摘要: |
| 为了普及国粹京剧，本研究提出一种将随机森林（Random Forest，RF）耦合注意力（Attention）机制和双向长短时记忆（BiLSTM）网络的音乐分类方法RF-BiLSTM-Attention，使用其进行京剧与其他类型音乐的二分类（以下简称“京剧二分类”）。首先，提取音乐所有光谱特征，利用RF选择重要特征；然后，在BiLSTM网络的隐藏层与输出层之间嵌入注意力层，对数据进行分类训练与预测。用来自大众音乐平台和GTZAN数据集的1 500首音乐进行京剧二分类实验，对比RF对循环神经网络（RNN）、长短时记忆（LSTM）网络、BiLSTM等9种模型的影响，结果表明：RF-BiLSTM-Attention模型的分类准确率为89.00%，运行时间为33.22 s，比简单模型中表现最好的RF-BiLSTM模型的分类准确率提高3.33%，运行时间缩短40.54%；比原始BiLSTM-Attention模型的分类准确率提高6.33%，运行时间缩短96.89%。与传统音频分类工作相比，本研究考虑了京剧二分类问题，对京剧起着良好的推广作用。 |
| 关键词: 京剧 双向长短时记忆网络 注意力机制 随机森林 二分类 |
| DOI：10.13657/j.cnki.gxkxyxb.20230829.012 |
| 投稿时间：2023-03-30修订日期：2023-07-04 |
| 基金项目:国家自然科学基金项目（12171378），冶金工业过程系统科学湖北省重点实验室项目（Y202105）和武汉科技大学研究生教学研究项目（Yjg202116）资助。 |
|
| Beijing Opera Binary Classification Simulation Based on RF-BiLSTM-Attention Music Classification Method |
|
GONG Yicheng1,2, LIU Qing1,2, XIAO Haoyi1
|
| (1.School of Science College, Wuhan University of Science and Technology, Wuhan, Hubei, 430065, China;2.Hubei Province Key Laboratory of Systems Science in Metallurgical Process (Wuhan University of Science and Technology), Wuhan, Hubei, 430081, China) |
| Abstract: |
| In order to popularize Beijing Opera, a music classification method of RF-BiLSTM-Attention was proposed based on Random Forest (RF) coupled Attention mechanism and Bidirectional Long Short Term Memory (BiLSTM) network, which was used to classify Beijing Opera and other types of music (hereinafter referred to as ‘Beijing Opera binary classification’).Firstly, all spectral features of music were extracted, and important features were selected by RF.Then, the attention layer was embedded the hidden layer and the output layer of the BiLSTM network to classify, train and predict the data.Using 1 500 pieces of music from the popular music platform and GTZAN dataset for Beijing Opera binary classification experiments, the effect of RF on 9 models such as Recurrent Neural Network (RNN), Long Short Term Memory (LSTM) and BiLSTM was compared.The results showed that the classification accuracy of RF-BiLSTM-Attention was 89.00%, with a run time of 33.22 s.Compared with the best performing RF-BiLSTM model in the simple model, the classification accuracy was improved by 3.33%, and the run time was reduced by 40.54%.Compared with the original BiLSTM-Attention model, the classification accuracy was increased by 6.33%, and the run time was shortened by 96.89%.Compared with traditional audio classification work, this article considers the binary classification of Beijing Opera, which plays a good role in promoting Beijing Opera. |
| Key words: Beijing Opera BiLSTM network attention mechanism random forest binary classification |
