Delayed Fusion: Integrating Large Language Models into First-Pass Decoding in End-to-End Speech Recognition
This paper presents an efficient decoding approach for end-to-end automatic speech recognition (E2E-ASR) with large language models (LLM). Although shallow ...