欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
为了解决语种识别中语音段长度失配以及短语音数据不充分带来的系统性能下降问题,提出了一种基于正则化的i-Vector改进算法.该算法通过对目标函数引入适当的正则化因子,构造新的目标函数进行优化,从而获得更好的i-Vector向量,提高解的稳定性.详细推导了正则化的目标函数构造过程和数学优化方法.语种识别实验证明,改进算法与基线系统相比,在测试语料为短语音段时,系统性能有一定的提升,测试语料越短,性能提升越明显.
英文摘要：
To alleviate the performance degradation of language recognition systems caused by the length mismatch among utterances and data sparseness for short speech segments, regularization based i-Vector estimation algorithm is proposed. A new objective function is constructed by introducing an appropriate regularization term in the likelihood function. After optimizing the objective function, a better i-Vector is obtained, which is more robust than that obtained by the original meth- od, especially in case of insufficient recognition data. This paper discusses how to introduce regular- ization term into the i-Vector algorithm, together with the corresponding optimization method in de- tail. Experimental results show that the regularization method improves the performance compared with that of the baseline system. The shorter the recognition segments, the more significant the per- formanee improvement.
同期刊论文项目
同项目期刊论文
期刊信息
位置：