一种基于PCA-BP 神经网络的示例优选方法

摘要

This paper proposes the example preferred method based on Principal Component Analysis(PCA)and BP Neural Network(BPNN)to solve computational cost, long retrieval time and poor noise robustness by the amount of example data in audio sample retrieval study. The paper builds the segment level features by principal component analysis, eliminates redundant data, and reduces the input variables, then models and predicts reserved ingredients by the BPNN. It tests the experimental data by the PCA-BPNN model confirmatory. Finally, experimental results show that the method can select optimization example from an audio accurately and efficiently.%在音频示例检索的研究中，针对示例数据量大而导致计算代价大、检索时间长和噪声鲁棒性差等问题，提出了一种基于主成分分析（PCA）和BP神经网络（BPNN）的示例优选方法。以信号鲁棒性评分为依据构建数据集合，使用主成分分析得到段级特征，消除数据冗余，减少输入变量，最后利用BPNN对保留成分进行建模预测。用PCA-BPNN模型对实验数据进行了验证性测试和分析，结果表明，该方法可以准确而高效地从一段音频中选取鲁棒性好的示例。