【摘要】 为施行有效的音频分类以高效率处理日渐复杂的音频信息,研究采用包含多种神经网络在内的5种机器学习模型,实现多种决策下的音频分类以寻找最优模型,基于分类准确度对各模型分类效果进行评估,在使用正则化方法保证模型泛化能力的条件下,通过比较和实验,挖掘并验证出了相对最优的模型——卷积神经网络音频分类模型及对应参数,为现有音频分类模型的进一步优化提供了参考方向。更多还原

【Abstract】 To implement effective audio classification for efficient audio information process which is becoming increasingly complex,five kinds of machine learning models including neural network were built respectively to find the best classification model,and the accuracy of classification of each model was evaluated.The relatively better model,convolutional neural network for audio classification,and its corresponding best parameters were recommended and verified based on experiment and comparison under dropout regulation to guarantee the ability of generalization.The new reference orientation was provided to optimize existing classification model.更多还原