Let's Fuse Step by Step: A Generative Fusion Decoding Algorithm with LLMs for Robust and Instruction-Aware ASR and OCR

TL;DR

This paper introduces Generative Fusion Decoding, a plug-and-play framework that integrates large language models into speech and text recognition systems, improving accuracy and adaptability without retraining.

Contribution

It presents a novel, cross-modal fusion decoding method that seamlessly combines LLMs with ASR and OCR, enabling better performance and instruction-aware adaptability.

Findings

Outperforms cascaded methods on English and Mandarin benchmarks

Achieves up to 17.7% WER reduction in adaptive ASR

Enables in-context learning transfer for improved recognition

Abstract

We propose "Generative Fusion Decoding" (GFD), a novel shallow fusion framework designed to integrate large language models (LLMs) into cross-modal text recognition systems for automatic speech recognition (ASR) and optical character recognition (OCR). We derive the necessary formulations to enable GFD to operate across mismatched token spaces of different models by calculating likelihood at the byte level, thereby enabling seamless fusion and synchronous progression during the decoding process. GFD is plug-and-play by design, making it readily compatible with various auto-regressive models without the need for any re-training. GFD proves effective for general ASR and OCR tasks through intermediate and frequent interactions with LLMs, surpassing cascaded methods in English and Mandarin benchmarks. In addition, GFD transfers in-context learning abilities of LLMs and allows for adaptive ASR in instruction-aware and long-context settings, yielding significant WER reductions of up to 17.7\%.