中科院声学所语音交互技术研究中心,北京,100080
纸质出版:2004
移动端阅览
罗 宇, 杜利民. 基于概率加权平均的Mel子带特征重建算法[J]. 电子学报, 2004,32(10):1738-1741.
LUO Yu, DU Li-min. Probability-Weighted Average Algorithm for Mel-Frequency Filter-Bank Vector Reconstruction[J]. Acta Electronica Sinica, 2004, 32(10): 1738-1741.
本文提出基于概率加权平均的Mel子带特征数据重建算法.该算法选择
K
个最优重建结果的概率加权平均作为被加性噪声掩蔽的语音特征分量的估计.实验结果表明
基于概率加权平均的语音特征数据重建算法降低了重建误差
减少了帧间突变现象
增强了Mel子带特征的帧间连续性
从而显著提高了语音识别系统对加性噪声的鲁棒性能.
In this paper
probability weighted average (PWA) algorithm is proposed to reconstruct Mel-frequency filter-bank vectors.The probability-weighted average of K-best reconstructed "missing" components of Mel-frequency filter-bank vectors is taken as the estimation of components masked by additive noise.Experimental results show that PWA algorithm can reduce reconstruction error
increase the continuity between neighbor mel-filterbank vectors and greatly improve automatic speech recognition (ASR) system's robustness against additive noise.
0
浏览量
976
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621