Speech Enhancement Based on Feature Reconstruction for Automatic Speech Recognition System with Unknown Structure

This study proposes a speech enhancement method which can be applied as a front-end to an automatic speech recognition system with an unknown structure. In this paper, a speech enhancement method is proposed, which is based on a feature reconstruction method employing variational model composition. In the proposed scheme, a gain parameter is estimated from the reconstructed speech feature and it is used for speech enhancement. The experimental results show that the proposed speech method significantly outperforms the existing front-end algorithms for unknown speech recognition over various background noise conditions. The results demonstrate that the proposed method can be effectively employed for an unknown ASR system to improve speech recognition performance, where no knowledge of the ASR system is available including the feature type and acoustic model.