Training Neural Speech Recognition Systems with Synthetic Speech Augmentation

TL;DR

This paper explores augmenting natural speech datasets with synthetic speech to improve neural speech recognition systems, achieving state-of-the-art results without external language models.

Contribution

It introduces a method of augmenting datasets with synthetic speech and demonstrates improved performance of large end-to-end neural ASR models.

Findings

Achieved state-of-the-art Word Error Rate on LibriSpeech

Synthetic speech augmentation enhances model accuracy

No external language model needed for top performance

Abstract

Building an accurate automatic speech recognition (ASR) system requires a large dataset that contains many hours of labeled speech samples produced by a diverse set of speakers. The lack of such open free datasets is one of the main issues preventing advancements in ASR research. To address this problem, we propose to augment a natural speech dataset with synthetic speech. We train very large end-to-end neural speech recognition models using the LibriSpeech dataset augmented with synthetic speech. These new models achieve state of the art Word Error Rate (WER) for character-level based models without an external language model.