Incorporating Language Level Information into Acoustic Models

TL;DR

This paper introduces Recurrent Deep Language Networks (RDLNs), a novel neural network architecture that integrates language-level information into acoustic models to enhance automatic speech recognition systems.

Contribution

The paper presents a new class of RDLNs that effectively incorporate language context into acoustic modeling, offering potential for improved ASR system fine-tuning.

Findings

RDLNs successfully integrate language context into acoustic models.

Multiple variants of RDLNs demonstrate flexible incorporation of language information.

Proposed methods improve the potential for end-to-end ASR system optimization.

Abstract

This paper proposed a class of novel Deep Recurrent Neural Networks which can incorporate language-level information into acoustic models. For simplicity, we named these networks Recurrent Deep Language Networks (RDLNs). Multiple variants of RDLNs were considered, including two kinds of context information, two methods to process the context, and two methods to incorporate the language-level information. RDLNs provided possible methods to fine-tune the whole Automatic Speech Recognition (ASR) system in the acoustic modeling process.