论文部分内容阅读
有效特征的选取一直都是语音情感识别算法的关键。为此,针对语音情感特征选择与构建的问题,一种仿选择性注意机制的语音情感识别算法被提出。考虑到语音信号的时频特性,算法首先计算语音信号的语谱图;其次,模仿选择性注意机制,计算语谱图的颜色、方向和亮度特征图,归一化后形成特征矩阵;然后,将特征矩阵重排列并进行PCA降维,形成情感识别特征向量;最后,利用改进的支持向量机分类方法进行语音情感识别。对愤怒、恐惧、高兴、悲伤和惊奇5种情感的识别实验显示,基于选择性注意的方法能够获得较好的识别效果,平均识别率为85.44%。相比于韵律特征和音质特征,语音情感识别率至少提高10%;相比于其它语谱特征,识别率提高7%左右。
The selection of effective features has always been the key to voice emotion recognition algorithm. Therefore, aiming at the problem of the selection and construction of speech emotion features, a speech emotion recognition algorithm imitating selective attention mechanism is proposed. Taking into account the time-frequency characteristics of the speech signal, the algorithm first calculates the speech spectrum of the speech signal. Secondly, it simulates the selective attention mechanism to calculate the color, direction and brightness features of the spectrogram and normalizes them to form the eigenmatrix. The feature matrix is rearranged and PCA dimensionality reduction is performed to form the emotion recognition feature vector. Finally, the speech recognition is improved by using the improved support vector machine classification method. Experiments on identifying five emotions of anger, fear, happiness, sadness and surprise show that the method based on selective attention can achieve better recognition results with an average recognition rate of 85.44%. Compared with the prosodic feature and the sound quality feature, the recognition rate of speech emotion is increased by at least 10%, and the recognition rate is improved by about 7% compared with other speech feature.