Domain Adaptation and Autoencoder-Based Unsupervised Speech Enhancement

As a category of transfer learning, domain adaptation plays an important role in generalizing the model trained one task and applying it to other similar tasks or settings. In speech enhancement, well-trained acoustic can be exploited obtain signal context languages, speakers, environments. Recent research was developed more effectively with various neural networks high-level abstract features. However, related studies are likely from rich diverse limited domain. Therefore, this study, method is proposed unsupervised enhancement for opposite circumstance that transferring larger richer On hand, importance-weighting (IW) approach variance constrained autoencoder reduce shift shared weights between source target domains. order train classifier worst-case minimize risk, minimax proposed. Both IW methods evaluated VOICE BANK IEEE datasets TIMIT dataset. The experiment results show outperform state-of-the-art approaches.