【摘要】 针对语音情感信号的复杂性和单一分类器识别的局限性,提出一种核函数极限学习机(KELM)决策融合的方法用于语音情感识别。首先对语音信号提取不同的特征,并训练相应的基分类器,同时将输出转化为概率型输出;然后利用测试集在基分类器的输出概率值计算自适应动态权值;最后对各基分类器的输出进行线性加权融合得到最终的分类结果。利用该方法对柏林语音库中4种情感进行识别,实验结果表明,提出的融合KELM方法优于常用的单分类器以及多分类器融合方法,有效地提高了语音情感识别系统的性能。更多还原

【Abstract】 In order to overcome the limitation of single classifier recognition and the complexity of emotional speech signal,a deci-sion fusion method based on Extreme Learning Machine with Kernel(KELM) is proposed for speech emotion recognition.Firstly,KELMs are built separtely by different features extracted from speech signal,while the outputs are transformed from numeric outputs into probability output.Then the confusion matrix of each classifier based on the test set is calculated,which is used to calculate the dynamic adaptive weight of the base classifiers.Finally the proposed method gets the ultimate classification result by linear weighted method.The fusion KELM is used to recognize four kinds of emotional speech in Berlin speech database,experimental re-sults show that the fusion KELM is superior to single classifier and multi classifier fusion method.It has improved the performance of the speech emotion recognition system effectively.更多还原