用功率谱相减法进行混噪语音识别的研究

Noisy Speech Recognition with Spectral Subtraction Methods

为了提高混噪语音的识别率,对以往功率谱相减法进行了数理解析,指出了其不足之处。针对某些混有频率成份较少的噪音的语音,提出了一种新的功率谱相减法,给出了相应的处理算法,并运用这种新的功率谱相减法进行了仿真语音识别实验,实验结果证明了这种新的功率谱相减法的正确性和有效性。

In order to upgrade noisy speech recognition, conventional spectral subtraction method is analyzed, the mistake of the method is pointed out. An algorithm used in new spectral subtraction method for noisy speech mixed few noise frequency components is proposed. Simulation experiment of speech recognition using the new spectral subtraction is carried out. The results demonstrate the correction and effectiveness of this new method.