人工合成噪声环境下的连续英文数字识别系统

噪声鲁棒性问题是当前语音识别的一个重点,作者提出了一个在已有数据库下通过人为地将噪声和语音信号混合的方法,实现实际环境下的连续英文数字语音识别系统.即通过自设计的程序将采集到的噪声文本根据不同的信噪比随机地添加到现有的语音数据库的语音文本中,使新的数据库中的语音文本符合实际的语音环境.实验结果表明,本系统对带噪声环境下的英文数字的识别率效果好,单词的总体正确识别率达到95.86%.

An English Digit Speech Recognition Under Articfical Noise Background

Now,how to improve the noisy robustness of system is the focus of the current speech recogni-tion.This paper put foward a new way to achieve the continues English d igit speech recognition system under artificial environment.It is by m ixing noise and speech signal artificially based on an existing speech database.In order to make the speech text in new database suitable for the actual noisy environ-ment,we random ly add the noise text to the existing speech text of speech database accord ing to the d if-ferent SNR by the program make by ourselves.The results demonstrate that the system recognize the Eng-lish d igit in noisy environment very well.It is found that the exactness of recognizing words is 95.86.