基于神经网络的语音情感识别

Research of speech emotion recognition based on acoustic features and ann

研究目的就是通过深入分析各种语音情感特征,找出其中对情感识别有较大贡献的特征,并寻找适合的模型将有效特征加以利用。分析和研究了多位科学家在进行语音情感分析过程中采用的方法和技术,通过总结和创新建立了语音情感语料库,并成功地提取了相关的语音信号的特征。研究了基音频率、振幅能量和共振峰等目前常用的情感特征在语音情感识别中的作用,重点研究了MFCC和△MFCC,实验发现特征筛选后系统的识别效果有着一定程度的提高。将处理后的频谱特征参数同原有的BP人工神经网络模型有效地结合起来,形成完整的语音情感识别系统,取得了较为满意的识别结果。

The main goal of this thesis is to search the most useful features with analyzing the features related with emotions, and find a recognition model to make use of these features.It studies the method and technology in the research of the speech emotion recognition,and creates the database of the speech emotion recognition and picks-up the features of the speech signal. Then it studies the effect in emotion-speech recognition from those common features such as pitch,amplitude energy,formant and so on.After choosing the useful features such as Mel-Frequency Cepstral Coefficients （MFCC） and its transient parameters,it obtains a better performance with the application of neural network.