改进卷积神经网络的音频场景分类研究

Research on acoustic scene classification based on improved convolutional neural network

音频场景分类作为声学场景理解的关键环节,对机器感知复杂环境并做出智能选择有着非常重要的意义。针对音频场景分类性能提升这一问题,提出改进的基于卷积神经网络模型的音频场景分类方法。首先对音频数据重新采样,预处理后得到对数梅尔谱图,随后输入到改进的卷积神经网络模型,进行卷积和池化处理提取谱图的特征,由Softmax分类器对音频场景标签进行分类。实验最后在城市音频数据集上进行十折交叉验证,实验结果表明,所提模型比传统的卷积神经网络模型的分类准确率更高,准确率达到了80%。

As a key part of acoustic scene understanding,acoustic scene classification is very important for the machine to perceive complex environments and make intelligent choices.In view of this,an acoustic scene classification method based on improved convolutional neural network model is proposed to enhance the acoustic scene classification performance.The audio data is subjected into resampling and preprocessing in sequence to get the log⁃mel spectrum.And then,the log⁃mel spectrum is input to the improved convolutional neural network model for convolution and pooling processing,so as to extract the spectrum features.Softmax classifier is used for the classification of audio scene labels.The 10⁃fold cross⁃validation was performed on the urban audio dataset.The experimental results show that the classification accuracy of the proposed classification method reaches 80%,which are higher than that of the traditional convolutional neural network model.