LSTM Language Models for LVCSR in First-Pass Decoding and Lattice-Rescoring

Abstract

LSTM based language models are an important part of modern LVCSR systems as they significantly improve performance over traditional backoff language models. Incorporating them efficiently into decoding has been notoriously difficult. In this paper we present an approach based on a combination of one-pass decoding and lattice rescoring. We perform decoding with the LSTM-LM in the first pass but recombine hypothesis that share the last two words, afterwards we rescore the resulting lattice. We run our systems on GPGPU equipped machines and are able to produce competitive results on the Hub5'00 and Librispeech evaluation corpora with a runtime better than real-time. In addition we shortly investigate the possibility to carry out the full sum over all state-sequences belonging to a given word-hypothesis during decoding without recombination.

References in corpus (2)

• The Microsoft 2017 Conversational Speech Recognition System
• The CAPIO 2017 Conversational Speech Recognition System

Cited by in corpus (9)

• Context-Dependent Acoustic Modeling without Explicit Phone Clustering
• Long-span language modeling for speech recognition
• LSTM-LM with Long-Term History for First-Pass Decoding in Conversational Speech Recognition
• Towards Consistent Hybrid HMM Acoustic Modeling
• The RWTH ASR System for TED-LIUM Release 2: Improving Hybrid HMM with SpecAugment
• Full-Sum Decoding for Hybrid HMM based Speech Recognition using LSTM Language Model
• On Sampling-Based Training Criteria for Neural Language Modeling
• A Parallelizable Lattice Rescoring Strategy with Neural Language Models
• Integrating Discrete and Neural Features via Mixed-feature Trans-dimensional Random Field Language Models