基于语义细胞的语音情感识别

摘要

Information cell was applied in the field of speech emotion recognition to address the problem of high space complexity of speech emotion recognition classifier . Single‐layered information cell (IC‐S ) algorithm and speaker‐emotion recognition based dual‐layered information cell (IC‐D ) algorithm were proposed in the light of information cell mixture model .Cross‐validation test on CASIA (in Chinese) and SAVEE (in English) corpus were conducted using F‐score as the indicator of recognition performance . Results show that the IC‐S algorithm has advantages in both time and space complexity compared to common algorithms like SVM .IC‐D algorithm achieves similar recognition performance as SVM . IC‐D algorithm can reduce the space complexity significantly and it is suitable for scenarios with few or fixed speakers .%为解决语音情感识别分类器空间复杂度高的问题，将语义细胞应用于语音情感识别领域。以语义细胞混合模型为核心，提出基于单层语义细胞的语音情感识别（IC‐S ）算法以及基于说话人‐情感识别的双层语义细胞识别（IC‐D）算法。在CASIA（汉语）和SAVEE（英语）情感语料库中进行交叉验证实验，并利用F值评判识别性能。结果表明：相比常用算法（如：SVM ），IC‐S算法在空间和时间复杂度上具有优势；IC‐D算法与SVM 算法识别准确率相似，可以有效降低模型存储空间的复杂度，适用于说话人分类较少或较为固定的场景。