A Discriminative Entity-Aware Language Model for Virtual Assistants

• DOI: 10.21437/interspeech.2021-1767
• Publication Date: 2021-08-27T05:59:39Z
• Authors: Mandana Saebi; Ernest Pusateri; Aaksha Meghawat; Christophe Van Gysel
• Keywords: Discriminative model; Entity linking; Knowledge graph; Word error rate

High-quality automatic speech recognition (ASR) is essential for virtual assistants (VAs) to work well. However, ASR often performs poorly on VA requests containing named entities. In this work, we start from the observation that many errors entities are inconsistent with real-world knowledge. We extend previous discriminative n-gram language modeling approaches incorporate knowledge a Knowledge Graph (KG), using features capture entity type-entity and entity-entity relationships. apply our model through an efficient lattice rescoring process, achieving relative sentence error rate reductions of more than 25% some synthesized test sets covering less popular entities, minimal degradation uniformly sampled set.