语音情感识别算法中新型参数研究

Practical Speech Emotion Recogniton Algorithm Research

语音情感识别是实现智能人机交互的关键技术之一。然而,用于语音情感识别的语音情感特征十分有限。为此,提出一种新型的语谱图显著性特征来改善语音情感识别效果。识别算法利用选择性注意模型获取语音信号语谱图像的显著图,并从中提取显著性特征,结合语音信号传统的时频特征构成语音情感识别特征向量。最后,利用KNN分类方法进行语音情感识别。实验结果表明,加入显著性特征后识别率有明显提升。

Speech Emotion recognition is one of the key technologies of intelligent human-computer interaction.However,the speech emotion feature for speech emotion recognition is very limited. Therefore,a new spectrogram of significant features is proposed to improve speech emotion recognition effect. Using selective attention model to obtain significant speech signal spectral image of the language,and extract significant features,recognition algorithm combined with the frequency characteristics of the speech signal constitutes the traditional speech emotion recognition feature vectors. Finally,we use KNN classification method for speech emotion recognition. Experimental results show that adding significant feature recognition rate has improved significantly.