基于人工神经网络的汉语数字语音识别系统

The Mandarin Digit Speech Recognition Based on Artificial Nerual Networks

关键词:  语音识别  人工神经网络  有序聚类算法  时间规整  孤立词

Key words:  speech recogrition  artificial nerual network  sequential cluster  time alignment  isolated word

摘要:

利用改进的有序聚类算法得到解决时间规整问题的新算法,在此基础上建立了基于人工神经网络的普通话数字语音识别系统。对基于人工神经网络的算法和基于动态时间伸缩的算法作比较识别实验,结果表明,基于人工神经网络的语音识别算法的识别性能优于传统的动态时间伸缩算法.

Abstract:

There is a time alignment problem in artificial neural network based speech recognition.In this paper,we adopt an improved sequential cluster method to solve the problem.A mandarin digit speech recognition system is established based on this method.The experiment results demonstrate that the improved time alignment method can improve the recognition rate effectively.