Can I read LLM Rescoring for ASR Systems Analysis on EtoBox?
LLM Rescoring for ASR Systems Analysis by gogigorgonzola is a document available to read on EtoBox.
What is LLM Rescoring for ASR Systems Analysis about?
This study investigates the impact of large-scale language model (LLM) rescoring on automatic speech recognition (ASR) systems, specifically using the Conformer-Transducer model. Results show that bidirectional LLMs like BERT and RoBERTa enhance ASR performance, while the unidirectional GPT-2 does not. Additionally, improvements are attributed to factors such as LLM pretraining, in-domain finetuning, and context augmentation.
- Author
- gogigorgonzola
- Language
- EN