(Mar 6) DB Augmentation in Acoustic Field Using Generative Model

Recently, performance enhancement has been reported in various fields through feature extraction based on deep learning and recognition structure. In recent years, performance enhancement has been studied by creating and reinforcing DB itself beyond features and recognizers. Especially, generative models based on deep learning are used for DB generation including Generative Adversarial Net (GAN). In order to confirm the applicability of this approach to the acoustic field, we will analyze a recent case where DB reinforcement is applied to the IEEE DCASE 2017 competition environment sound recoginition, moreover a method for verifying the generated DB is also discussed.