Transformer-based language modeling and decoding for conversational speech recognition (2001.01140v1)

Abstract: We propose a way to use a transformer-based LLM in conversational speech recognition. Specifically, we focus on decoding efficiently in a weighted finite-state transducer framework. We showcase an approach to lattice re-scoring that allows for longer range history captured by a transfomer-based LLM and takes advantage of a transformer's ability to avoid computing sequentially.