摘要
本文提出了一种提高不匹配环境下的语音识别性能的方法。该方法基于多层感知机映射，采用最大似然一后向传播（ML－BP）算法，利用少量测试环境下的语音进行训练，实现训练环境与测试环境的语音特征空间或模型空间之间的映射。实验证明，采用了本文提出的算法后，测试环境下的语音识别性能有了很大的提高。
This paper proposes a robust speech recognition algorithm which is based on the Multi-Layer Perceptron. Inthis method the speech samples under the mismatched environment are trained with the ML-BP algorithm to realize themapping between the feature spaces or the model spaces of the training environment and the testing environment.Experiments demonstrate that the robustness of the speech recognition under the testing environments can be considerablyimproved.
出处
《电路与系统学报》
CSCD
1999年第4期14-18,共5页
Journal of Circuits and Systems
基金
国家自然科学基金!69772020
国家"863"项目!863-512-9805-10
