基于传统的Mel倒谱系数(MFCC)系列特征的语音识别系统在噪声环境中的识别性能会急剧下降。为了进行噪声环境中的自动语音识别,提出了一种反映语音信号谐振程度的特征:谐振强度,并用之代替传统MFCC特征中的能量维(零维倒谱C0,或者帧能量E)。在展览馆噪声、人群噪声和汽车噪声等情况下的语音识别实验结果表明:基于这种新特征的语音识别系统比基于传统特征的语音识别系统有更高的平均识别率和更好的抗噪声能力。 Speech recognition systems based on the traditional Mel-Cepstral Coefficients (MFCC) series feature a sharp drop in recognition performance in noisy environments. In order to carry out automatic speech recognition in noisy environments, a new characteristic, resonant strength, is proposed to represent the resonance of speech signals, and used instead of the energy dimension (zero-dimensional cepstrum C0, or frame energy E) in traditional MFCC features. The results of speech recognition in the exhibition hall noise, crowd noise and car noise show that the speech recognition system based on this new feature has higher average recognition rate and better anti-noise than the speech recognition system based on traditional features ability.