论文部分内容阅读
为了提高语音情感识别率,提出一种新的特征融合方法.在全局特征的基础上,利用各种不同特征的局部信息,把全局特征和局部特征结合起来,引入多核学习的方法,使整体的全局特征和每类局部特征都对应一个核函数,加权求和得到一个组合核进行非线性映射,使不同类别的情感特征在高维再生核Hilbert空间中变得更容易分开.采用Berlin语音情感数据库,利用交叉验证的方法确定相应的全局核和局部核的参数,经过多核学习计算,得到所有核的权重,确定共振峰和强度是情感识别中相对重要的特征.实验表明,采用传统的方法识别率为78.74%,而采用所提出的方法,识别率为81.10%.因此,所提出的特征融合方法能够有效地提高语音情感的识别率.
In order to improve speech emotion recognition rate, a new feature fusion method is proposed based on global features, using local information of various features to combine global features and local features, and to introduce multi-core learning method so that the whole The global features and each type of local features correspond to a kernel function, and the weighted summation results in a combined kernel for nonlinear mapping, making it easier for different categories of affective features to be separated in the Hilbert space of high dimensional regenerative kernels. Using the Berlin Speech Emotional Database , Cross-validation method to determine the corresponding global nuclear and local nuclear parameters, through the calculation of multi-core learning to get all the nuclear weights, to determine the resonance peak and intensity of emotional recognition is a relatively important feature.Experiments show that the use of traditional methods to identify The rate is 78.74%, and the recognition rate is 81.10% using the proposed method.Therefore, the proposed feature fusion method can effectively improve the recognition rate of speech emotion.