摘要
实际应用中与文本无关的说话人识别研究,模型训练的说话人语音一般是有限的。此外,由于说话人自身生理因素的改变、外部采集环境的变化等都可能会导致说话人语音的声学特征发生改变。因此,代表说话人模型的特征分布也在不断变化,从而造成说话人识别系统识别率下降。文中在说话人自适应技术的基础上,提出了说话人模型的连续自适应算法,解决了因说话人自身声学特征的变化导致识别率下降的问题。
In practical application,the speaker training data is generally limited for the study of text-independent speaker recognition.In addition,changes such as the speaker's own physiological factor and the channel transmission characteristics may cause the change of acoustic characteristics of speaker's speech.Therefore,the distribution on behalf of the speaker model is also changed constantly,which leads to the decrease of the recognition rate of speaker recognition system.In this paper,based on speaker adaptation techniques,the continues adaptive algorithm of speaker model is presented in order to solve the problem of the decline of the recognition rate due to the change of speaker's acoustic characteristics.
出处
《通信电源技术》
2016年第2期81-83,共3页
Telecom Power Technology
基金
贵州省社会发展攻关项目(黔科合SY字[2013]3105号)
