论文部分内容阅读
本文从模板匹配的角度研究了多层前向感知机(MLP)在汉语孤立数目字语音识别中的应用,针对训练样本数受限的情况提出了新的训练方法,研究了语音固化、特征提取、学习算法和策略诸方面问题。对特定人和非特定人汉语孤立数目字语音识别分别达到了95.7%和93.0%(无拒识)的识别率。
In this paper, we study the application of multi-layer forward perceptron (MLP) in Chinese isolated digital speech recognition from the perspective of template matching. We propose a new training method based on the limited number of training samples, and study the speech curing, feature extraction, Learning algorithms and strategies in all aspects of the problem. The recognition rate of 95.7% and 93.0% (without rejection) of Chinese isolate number words for specific persons and non-specific persons respectively.