| VTS特征补偿算法在语音识别中的实用性研究 |
| |
| 引用本文： | 杨钊,杜俊,胡郁,刘庆峰,戴礼荣.VTS特征补偿算法在语音识别中的实用性研究[J].小型微型计算机系统,2011,32(4). |
| |
| 作者姓名： | 杨钊 杜俊 胡郁 刘庆峰 戴礼荣 |
| |
| 作者单位： | 中国科学技术大学讯飞语音实验室,安徽,合肥,230027 |
| |
| 摘 要： | 在语音识别实际应用中,由于噪声的多样性,会造成训练和测试的失配,导致系统性能下降.特征补偿作为鲁棒性语音识别的一种重要方法,通过对训练和测试环境之间差异的研究,在特征空间中修正语音特征,使得修正后的测试语音特征能够更加接近训练语音特征.本文介绍一种实用的基于环境模型矢量泰勒级数(VTS)近似的特征补偿算法.首先验证传统的VTS离线算法在实际车载环境下的有效性;其次由于离线算法本身运算量很大,为了使其实用化,本文对算法进行改进,使其在提高效率的同时又能够保证与离线时相当的性能.通过实验结果验证,本文提出的实用化VTS算法在识别性能上相当接近离线时最好的性能.
|
| 关 键 词： | 失配 矢量泰勒级数 实用化 特征补偿 |
Application of VTS Approximation Based Feature Compensation Approach to Speech Recognition |
| |
| YANG Zhao,DU Jun,HU Yu,LIU Qing-feng,DAI Li-rong.Application of VTS Approximation Based Feature Compensation Approach to Speech Recognition[J].Mini-micro Systems,2011,32(4). |
| |
| Authors: | YANG Zhao DU Jun HU Yu LIU Qing-feng DAI Li-rong |
| |
| Affiliation: | YANG Zhao,DU Jun,HU Yu,LIU Qing-feng,DAI Li-rong(iFly Speech Lab,University of Science and Technology of China,Hefei 230027,China) |
| |
| Abstract: | In the environment with various noises,the mismatch between training and testing will result in a great reduction of speech recognition performance.As an effective method for robust speech recognition,feature compensation can refine testing noisy feature to be closer to training feature.In this paper,we proposed a practical feature compensation approach based on Vector Taylor Series(VTS) approximation using explicit model of environmental distortion.Firstly,we test and verify the effectiveness of traditiona... |
| |
| Keywords: | mismatch vector taylor series practical feature compensation |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
