Soft decisions in missing data techniques for robust automatic speech recognition
Soft decisions in missing data techniques for robust automatic speech recognition
复制标题
DOI：
10.21437/icslp.2000-92
复制
发表时间：
2000-10
期刊：
影响因子：
--
通讯作者：
J. Barker;Ljubomir Josifovski;M. Cooke;P. Green
中科院分区：
文献类型：
--
作者：
J. Barker;Ljubomir Josifovski;M. Cooke;P. Green
在以前的工作中，我们已经发展了这一理论，并展示了丢失数据方法在稳健自动语音识别中的前景。这种技术是基于关于每个时间-频率\像素“是可靠还是不可靠”的硬判决。在本文中，我们用每个\像素“可靠”的概率的软估计来代替这些离散判决。我们调整概率计算以使用这些估计作为每个特征向量分量的互补可靠/不可靠解释的加权因子。使用TIDI-GITS连接数字识别任务的实验表明，该方法在低信噪比下不能改善性能。
In previous work we have developed the theory and demonstrated the promise of the Missing Data approach to robust Automatic Speech Recognition. This technique is based on hard decisions as to whether each time-frequency \pixel" is either reliable or unreliable. In this paper we replace these discrete decisions with soft estimates of the probability that each \pixel" is reliable. We adapt the probability calculation to use these estimates as weighting factors for the complementary reliable/unreliable interpretations for each feature vector component. Experiments using the TIDi-gits connected digit recognition task demonstrate that this technique a(cid:11)ords signi(cid:12)cant performance improvements at low SNRs.