Updated Corpora and Benchmarks for Long-Form Speech Recognition

TL;DR

This paper updates popular ASR corpora for long-form speech recognition, investigates train-test mismatch issues, especially affecting AEDs, and demonstrates that long-form training improves model robustness in real-world scenarios.

Contribution

It re-releases key ASR datasets with updated annotations for long-form speech research and benchmarks training methods to address domain shift issues.

Findings

AEDs are more affected by train-test mismatch than transducers.

Long-form training improves robustness to domain shift.

Updated corpora facilitate research on real-world long-form speech recognition.

Abstract

The vast majority of ASR research uses corpora in which both the training and test data have been pre-segmented into utterances. In most real-word ASR use-cases, however, test audio is not segmented, leading to a mismatch between inference-time conditions and models trained on segmented utterances. In this paper, we re-release three standard ASR corpora - TED-LIUM 3, Gigapeech, and VoxPopuli-en - with updated transcription and alignments to enable their use for long-form ASR research. We use these reconstituted corpora to study the train-test mismatch problem for transducers and attention-based encoder-decoders (AEDs), confirming that AEDs are more susceptible to this issue. Finally, we benchmark a simple long-form training for these models, showing its efficacy for model robustness under this domain shift.