调制域谱减法用于鲁棒性语音识别

针对语音识别在实际环境中缺乏稳健性的问题,提出了将调制域谱减法应用于语音识别前端的方法。先将语音信号变换到调制域,在调制域用谱减法将信号增强,在减少语音畸变的基础上提高信噪比,然后再进行识别。仿真实验表明,调制域谱减法能在较大的信噪比区间内提高系统识别率,证明此方法能显著提高语音识别系统的抗噪声能力。

Modulation domain spectral subtraction forrobust speech recognition

In order to solve the problem that the rate of speech recognition will met a sharp decline in real environments , a modulation domain spectral subtraction method was proposed. This method let the time domain speech signal changed to modulation domain first, then use the spectral subtraction method to realized enhancement operation. This process can improved the SNR of speech effectly and reduce the distortion at the same time, finally used the processd speech in the recognition system. The experiment results showed that the proposed method got a good recognition rate promote in a wide SNR range. It means that the noise immunity of the speech recognition system has been improved well by using this method.