Long-span language modeling for speech recognition

Abstract

We explore neural language modeling for speech recognition where the context spans multiple sentences. Rather than encode history beyond the current sentence using a cache of words or document-level features, we focus our study on the ability of LSTM and Transformer language models to implicitly learn to carry over context across sentence boundaries. We introduce a new architecture that incorporates an attention mechanism into LSTM to combine the benefits of recurrent and attention architectures. We conduct language modeling and speech recognition experiments on the publicly available LibriSpeech corpus. We show that conventional training on a paragraph-level corpus results in significant reductions in perplexity compared to training on a sentence-level corpus. We also describe speech recognition experiments using long-span language models in second-pass re-ranking, and provide insights into the ability of such models to take advantage of context beyond the current sentence.

References in corpus (4)

• Training Tips for the Transformer Model
• Language Modeling with Deep Transformers
• Notes on Noise Contrastive Estimation and Negative Sampling
• LSTM Language Models for LVCSR in First-Pass Decoding and Lattice-Rescoring

Cited by in corpus (5)

• Deep Transfer Learning for Automatic Speech Recognition: Towards Better Generalization
• Adapting Long Context NLM for ASR Rescoring in Conversational Agents
• Cross-Utterance Language Models with Acoustic Error Sampling
• Transformer Language Models with LSTM-based Cross-utterance Information Representation
• Leveraging Cross-Utterance Context For ASR Decoding