摘要
音频信号的提取和分类在语音识别、语音合成、音乐分析等领域中具有重要的应用。而传统方法在对音频进行提取和分类中仍存在一定的局限性。因此,研究尝试以深度学习算法为基础,首先对音频的特征参量提取进行了步骤优化,提出了一种特征参量组合提取法。另外引入模拟退火算法对反向传播神经网络进行优化,联合特征参量提取方法后提出了一种新型音频特征分类模型。实验结果表明,特征参量组合提取法的特征项被选中次数最多为45次,此时权值最高为打击乐器中0.044。分类模型的准确度最高为90%,音频分类的平均用时最少为3.5 s。由此可知,研究所提方法在诸多的现有方法中存在显著优势,能够为音频信号处理提供一种新的解决方案。
The extraction and classification of audio signals have important applications in speech recognition,speech synthesis,music analysis and other fields.And the traditional methods still have some limitations in extracting and classifying audio.Therefore,the study tries to optimize the steps of feature parameter extraction of audio based on deep learning algorithm,and proposes a feature parameter combination extraction method.In addition,the simulated annealing algorithm is introduced to optimize the backpropagation neural network,and a new audio feature classification model is proposed after the combined feature parameter extraction method.The experimental results show that the feature items of the combined feature parameter extraction method are selected up to 45 times,when the highest weight value is 0.044 in percussion instruments.The accuracy of the classification model was up to 90%and the average time taken for audio classification was a minimum of 3.5 seconds.It can be seen that the proposed method of the study has significant advantages among many existing methods and can provide a new solution for audio signal processing.
作者
柴磊
CHAI Lei(Shaanxi Polytechnic Institute,Xianyang Shaanxi 712000,China)
出处
《自动化与仪器仪表》
2024年第12期196-199,204,共5页
Automation & Instrumentation
基金
《高职院校以美育课程为载体,培养时代工匠的探索与实践》(SZ21B043)。
关键词
深度学习
音频信号
特征参量
反向传播神经网络
音乐分类
deep learning
audio signal
feature covariates
backpropagation neural network
music classification
