基于深度学习可变长度语音片段的情感识别

Emotion Recognition of Variable-length Speech Segments Based on Deep Learning

通过将深度神经网络直接应用于频谱图,提出了一种用于可变长度语音段的情感识别方法。频谱图包含对情绪识别有用的对话语言信息。从频谱图中提取这些信息,并通过将卷积神经网络(CNN)与递归神经网络(RNN)相结合来完成情感识别任务。与传统的将句子分割成更小的固定长度段的方法相比,该方法可以解决语音分割过程中引入的准确性降低问题。实验结果表明,该方法在加权精度(WA)和不加权精度(UA)上均优于定长神经网络。

An approach of emotion recognition was proposed in this paper for the variable-length speech segments by applying deep neutral network to spectrograms directly.The spectrogram carries para-lingual information that is useful for emotion recognition.The information was extracted from spectrograms and the emotion recognition task was accomplished by combining Convolutional Neural Networks(CNNs)with Recurrent Neural Networks(RNNs).Compared to the traditional methods that split the sentence into smaller fixed-length segments,the method can solve the problem of accuracy degradation introduced in the speech segmentation process.Experimental results demonstrate that the proposed method outperforms the fixed-length neural network on both weighted accuracy(WA)and unweighted accuracy(UA).