End-to-End Open Vocabulary Keyword Search

• DOI: 10.21437/interspeech.2021-1399
• Publication Date: 2021-08-27T05:59:39Z
• Authors: Bolaji Yusuf; Alican Gok; Batuhan Gundogdu; Murat Saraclar
• Keywords: End-to-end principle

Recently, neural approaches to spoken content retrieval have become popular.However, they tend be restricted in their vocabulary or ability deal with imbalanced test settings.These restrictions limit applicability keyword search, where the set of queries is not known beforehand, and system should return just whether an utterance contains a query but exact location any such occurrences.In this work, we propose model directly optimized for search.The takes as input returns sequence probabilities each frame having occurred that frame.Experiments show proposed only outperforms similar end-to-end models on task ratio positive negative trials artificially balanced, it also able far more challenging search its inherent imbalance.Furthermore, using our rescore outputs LVCSR-based leads significant improvements latter.