Fast offline Transformer-based end-to-end automatic speech recognition for real-world applications

TL;DR

This paper introduces a fast, Transformer-based end-to-end speech recognition system optimized for real-world applications, achieving rapid processing of large speech datasets with high accuracy and efficiency.

Contribution

It proposes novel techniques to accelerate Transformer-based speech recognition, including multi-utterance batched decoding, end-of-speech detection with CTC, and speech segmentation, demonstrating significant speed improvements.

Findings

Recognizes 8 hours of speech in under 3 minutes.

Achieves a 10.73% character error rate on real-world data.

Reduces recognition time by 27.1% compared to conventional systems.

Abstract

With the recent advances in technology, automatic speech recognition (ASR) has been widely used in real-world applications. The efficiency of converting large amounts of speech into text accurately with limited resources has become more important than ever. This paper proposes a method to rapidly recognize a large speech database via a Transformer-based end-to-end model. Transformers have improved the state-of-the-art performance in many fields. However, they are not easy to use for long sequences. In this paper, various techniques to speed up the recognition of real-world speeches are proposed and tested, including decoding via multiple-utterance batched beam search, detecting end-of-speech based on a connectionist temporal classification (CTC), restricting the CTC prefix score, and splitting long speeches into short segments. Experiments are conducted with the Librispeech English and the real-world Korean ASR tasks to verify the proposed methods. From the experiments, the proposed system can convert 8 hours of speeches spoken at real-world meetings into text in less than 3 minutes with a 10.73% character error rate, which is 27.1% relatively lower than that of conventional systems.