摘要
针对说话人的语音特征和说话人的个性特征很难分离的问题,提出了一种基于遗传算法和支持向量机的说话人辨认新方法,再结合特征各分量的相对重要性,实现了与文本无关的说话人辨认系统。采用30维特征,识别率从97.45%提高到了97.81%,实验表明GA—SVM加重算法提取的特征对系统有更好的识别能力,能从大量语音特征中提取出说话人的个性特征。
A new method,based on genetic algorithm(GA)and support vector machines(SVMs),is proposed for speaker identification.And then combining the terms relative importance achieves a text-independent speaker identification system.The main difficulty is hard to isolate speakers' speech character and individuality character.Practical results show that the compound features generated by GA-SVM possess better recognition ability than the initial features do.
出处
《电子技术（上海）》
2008年第2期52-53,共2页
Electronic Technology
关键词
说话人辨认
支持向量机
遗传算法
Speaker Identification
Support Vector Machine
Genetic Algorithms
