Speech recognition improves by merging large language models efficiently
Merging the Knowledge of LLMs for Automatic Speech Recognition
Computation and Language
Summary
Speech-to-text systems often need extra language knowledge to better understand sentences, but using separate language models can slow them down. The authors propose a way to combine these language models directly into the main speech recognition model so it runs just as fast. They tested their method on different speech datasets and found it improved accuracy without making the system slower or use more memory.
What this means in practice
- •For voice assistant developers: Integrate multiple language models directly into speech recognition to improve accuracy without added runtime costs on smart devices.
- •For call center system engineers: Enhance speech-to-text transcription quality in specific domains by merging adapted language info without slowing down processing.
Authors
Hayato Futami, Tatsuya Kawahara
Abstract
Automatic speech recognition (ASR) systems, trained on paired speech-text data, have been improved by leveraging language models (LMs) trained on text-only data. LM fusion methods such as shallow fusion and density ratio are well-established methods that incorporate external LMs during ASR decoding. However, they incur additional computational costs due to LM inference, which is particularly problematic for recent larger LMs. In this study, we propose incorporating external LMs via model merging. This method integrates the LMs directly into the parameters of an LLM-based ASR model, requiring no additional computational cost at inference. We formulate domain extension and transfer via arithmetic operations on LoRA parameters. Experimental evaluations were conducted for the domain adaptation of LLM-based ASR trained on CSJ and LibriSpeech. We show that our LM merging consistently improved the ASR performance in the target domains, without degrading inference speed or memory footprint.