Integrating Text Inputs For Training and Adapting RNN Transducer ASR Models
This addresses the problem of domain adaptation for end-to-end ASR systems, enabling customization with only text data, which is incremental but offers practical gains.
The paper tackles the difficulty of customizing end-to-end ASR models by proposing a novel text representation and training framework, showing that an RNN Transducer model's internal LM component can be adapted with text-only data, resulting in up to 45% relative WER reduction on domain-specific datasets.
Compared to hybrid automatic speech recognition (ASR) systems that use a modular architecture in which each component can be independently adapted to a new domain, recent end-to-end (E2E) ASR system are harder to customize due to their all-neural monolithic construction. In this paper, we propose a novel text representation and training framework for E2E ASR models. With this approach, we show that a trained RNN Transducer (RNN-T) model's internal LM component can be effectively adapted with text-only data. An RNN-T model trained using both speech and text inputs improves over a baseline model trained on just speech with close to 13% word error rate (WER) reduction on the Switchboard and CallHome test sets of the NIST Hub5 2000 evaluation. The usefulness of the proposed approach is further demonstrated by customizing this general purpose RNN-T model to three separate datasets. We observe 20-45% relative word error rate (WER) reduction in these settings with this novel LM style customization technique using only unpaired text data from the new domains.