基于MFCC的语音情感特征提取研究

发布时间：2018-07-25 13:34

【摘要】：为了研究"世界记忆遗产"东巴经典古籍的音频分类,以通过语音情感特征提取的方法分类鉴别东巴音频类别,并实现对东巴经典语音的情感状态识别,同时提高人机交互性能,提出采用Mel频率倒谱系数(MFCC)实现语音情感特征的提取。通过引入MFCC的一阶差分、二阶差分描述语音特征的动态特征,并整合短时能量特征,最终形成MFCC和短时能量相叠加的语音信号特征参数,达到提取反映语音情感特征的目的。实验验证表明,该语音信号特征提取方法能够更明显地区分出包含在语音中的情感信息,为语音情感特征的识别研究及东巴古籍音频分类鉴别提供理论基础。
[Abstract]:In order to study the audio classification of the classic ancient books of the "World Memory Heritage" of Dongba, the audio classification of Dongba was identified by the method of speech emotion feature extraction, and the emotional state recognition of the classic voice of Dongba was realized and the human computer interaction performance was improved. The Mel frequency cepstrum coefficient (MFCC) was used to extract the emotion feature of the speech. After introducing the first order difference of MFCC, the two order difference describes the dynamic characteristics of the speech feature, and integrates the short time energy characteristics, and finally forms the feature parameters of the speech signal which is superimposed by the MFCC and the short-time energy, and achieves the purpose of extracting the emotional characteristics of the speech. The experimental verification shows that the speech signal feature extraction method can be more clearly divided in the region. The emotion information contained in speech provides a theoretical basis for the recognition of speech emotion characteristics and the audio classification and identification of Dongba ancient books.
【作者单位】：北京信息科技大学机电学院;中央民族大学;丽江东巴文化研究院;
【基金】：国家社科基金重大项目(12&ZD234) 现代测控技术教育部重点实验室开放课题(KF20161123205) 北京市重点实验室开放课题(KF20161123208)资助项目
【分类号】：TN912.3

【相似文献】