基于CNN的普米语孤立词语谱图分类

Primi Isolated Word Spectrogram Classification Basedon Convolutional Neural Network

为实现普米语孤立词语谱图的分类,引入基于卷积神经网络的语谱图模型,该模型可以无监督学习语谱图特征实现分类.本文搭建了一个9层的卷积神经网络模型,利用彩色语谱图样本集进行训练,并针对已训练好的模型,通过实验检验各项因素对分类的影响,从而得到适当的参数.参数确定后,进行卷积神经网络与支持向量机、BP神经网络的对比实验,验证算法的可行性和有效性.实验显示基于卷积神经网络的普米语孤立词语谱图分类准确率达到91%~95%,这说明该算法是可行和有效的.与支持向量机、BP神经网络相比,卷积神经网络具有自动提取特征,避免过拟合问题,适合大样本数据进行训练的优点.

A spectrogram model based on convolutional neural networks is introduced to achieve the classification of Primi isolated words.This model can achieve classification through unsupervised learning of spectrogram features.A nine-layer convolutional neural network modelis built,which is trained using the spectrogram sample set.For the trained model,the influence of various factors on the experimental results is tested by experiments,and thus the appropriate parameters are obtained.After the parameters are determined,a comparative experiment of convolutional neural network,support vector machine and BP neural network is performed,and the results show that the accuracy of Primi isolated word spectrogram classification based on convolutional neural networkis as high as 91%~95%,thus indicating its effectiveness and feasibility.Compared with SVM and BP neural network,this convolutional neural network has the advantages of automatically extracting features,avoiding over-fitting,and adapting to large sample data for training.