Unsupervised Multi-Target Domain Adaptation for Acoustic Scene Classification

Abstract

It is well known that the mismatch between training (source) and test (target) data distribution will significantly decrease the performance of acoustic scene classification (ASC) systems. To address this issue, domain adaptation (DA) is one solution and many unsupervised DA methods have been proposed. These methods focus on a scenario of single source domain to single target domain. However, we will face such problem that test data comes from multiple target domains. This problem can be addressed by producing one model per target domain, but this solution is too costly. In this paper, we propose a novel unsupervised multi-target domain adaption (MTDA) method for ASC, which can adapt to multiple target domains simultaneously and make use of the underlying relation among multiple domains. Specifically, our approach combines traditional adversarial adaptation with two novel discriminator tasks that learns a common subspace shared by all domains. Furthermore, we propose to divide the target domain into the easy-to-adapt and hard-to-adapt domain, which enables the system to pay more attention to hard-to-adapt domain in training. The experimental results on the DCASE 2020 Task 1-A dataset and the DCASE 2019 Task 1-B dataset show that our proposed method significantly outperforms the previous unsupervised DA methods.

On The Impact of Normalization Strategies in Unsupervised Adversarial Domain Adaptation for Acoustic Scene Classification

“…Despite its effectiveness, it requires a large number of recordings from the target device to carry out the adaptation process. Furthermore, the adaptation process could implicitly benefit from the parallel data present in the dataset, or explicitly use these data to ease generalization [16,17]. To overcome these limitations, band-wise statistical matching (BSWM) was introduced as a simple, linear UDA method for ASC that does not require any adaptation stage [18,19].…”