Domain Attentive Fusion for End-to-end Dialect Identification with\n Unknown Target Domain

End-to-end deep learning language or dialect identification systems operate\non the spectrogram or other acoustic feature and directly generate\nidentification scores for each class. An important issue for end-to-end systems\nis to have some knowledge of the application domain, because the system can be\nvulnerable to use cases that were not seen in the training phase; such a\nscenario is often referred to as a domain mismatched condition. In general, we\nassume that there is enough variation in the training dataset to expose the\nsystem to multiple domains. In this work, we study how to best make use a\ntraining dataset in order to have maximum effectiveness on unknown target\ndomains. Our goal is to process the input without any knowledge of the target\ndomain while preserving robust performance on other domains as well. To\naccomplish this objective, we propose a domain attentive fusion approach for\nend-to-end dialect/language identification systems. To help with\nexperimentation, we collect a dataset from three different domains, and create\nexperimental protocols for a domain mismatched condition. The results of our\nproposed approach, which were tested on a variety of broadcast and YouTube\ndata, shows significant performance gain compared to traditional approaches,\neven without any prior target domain information.\n