基于深度神经网络的电子音乐音质评估研究  

Research on Electronic Music Quality Evaluation Based on Deep Neural Network

作　　者：刘颖 LIU Ying(College of Music,Xianyang Normal University,Xianyang 712000,China)

机构地区：[1]咸阳师范学院,音乐学院,陕西咸阳712000

出　　处：《微型电脑应用》2022年第10期142-145,共4页Microcomputer Applications

摘　　要：为了降低电子音乐音质评估结果误差,提出基于深度神经网络的电子音乐音质评估方法。利用音频采样、归一化、分帧以及时频与变换等过程完成电子音乐预处理,确定电子音乐浊音段;在此基础上利用维特比算法跟踪浊音段主导基频轨迹,同时利用基频判别模型确定电子音乐主旋律。分析电子音乐主旋律内影响音质的声源特性、音响器材的信号特性、声场特性、听觉特性和立体感等类型的影响因子,引入对照样本,生成样本集。以开源项目Keras人工神经网络库为基础构建由输入层、归一化层、全连接层、激活层共同组成的深度神经网络,将样本集内数据作为输入,通过训练完成电子音乐音质评估。实验结果显示,该方法可准确提取电子音乐主旋律,且所构建模型的AUC值较高。In order to reduce the error of electronic music quality evaluation results,this paper studies the method of electronic music quality evaluation based on deep neural network.Based on the electronic music model,the music pitch can be determined by the electronic music model.This paper analyzes the influence factors of sound source characteristics,sound equipment signal characteristics,sound field characteristics,hearing characteristics and stereo sense,which affect the sound quality in the main melody of electronic music,and introduces control samples to generate sample sets.Based on the open source project Keras artificial neural network library,a deep neural network composed of input layer,normalization layer,full connection layer and activation layer is constructed.The data in the sample set are used as input,and the electronic music sound quality evaluation is completed through training.The experimental results show that the method can accurately extract the main melody of electronic music,and the AUC value of the model is high.

关 键 词：深度神经网络 电子音乐 音质评估 浊音段 基频轨迹 

分 类 号：TM933[电气工程—电力电子与电力传动]