A Log-Energy Feature Normalization Method Using ARMA Filter

摘要: The difference of environments between training and recognition is the major reason degradation speech recognition. To solve this mismatch environments, various noise processing methods have been studied. Among them, ERN(log-Energy dynamic Range Normalization) SEN(Silence Energy for normalization log energy features show better performance than others. However, these a problem that they can hardly achieve relatively higher values environmental caused by becomes bigger especially in low SNR environments. problems, we propose applying ARMA filter as post-processing smoothing calculating moving average auto-regression scheme. From results conducted on Aurora 2.0 DB, proposed method shows improved comparing with conventional methods.