Innovative Bert-based Reranking Language Models for Speech Recognition

TL;DR

This paper introduces BERT-based models for reranking speech recognition hypotheses, leveraging contextual understanding and global topic information to improve accuracy, demonstrated through experiments on the AMI corpus.

Contribution

It presents a novel BERT-based reranking approach (PBERT and TPBERT) for speech recognition, incorporating global topic information for enhanced hypothesis selection.

Findings

PBERT outperforms traditional RNN-based reranking.

TPBERT effectively utilizes global topic information.

Methods show significant WER reduction on AMI corpus.

Abstract

More recently, Bidirectional Encoder Representations from Transformers (BERT) was proposed and has achieved impressive success on many natural language processing (NLP) tasks such as question answering and language understanding, due mainly to its effective pre-training then fine-tuning paradigm as well as strong local contextual modeling ability. In view of the above, this paper presents a novel instantiation of the BERT-based contextualized language models (LMs) for use in reranking of N-best hypotheses produced by automatic speech recognition (ASR). To this end, we frame N-best hypothesis reranking with BERT as a prediction problem, which aims to predict the oracle hypothesis that has the lowest word error rate (WER) given the N-best hypotheses (denoted by PBERT). In particular, we also explore to capitalize on task-specific global topic information in an unsupervised manner to assist PBERT in N-best hypothesis reranking (denoted by TPBERT). Extensive experiments conducted on the AMI benchmark corpus demonstrate the effectiveness and feasibility of our methods in comparison to the conventional autoregressive models like the recurrent neural network (RNN) and a recently proposed method that employed BERT to compute pseudo-log-likelihood (PLL) scores for N-best hypothesis reranking.