Is pushing numbers on a single benchmark valuable in automatic speech recognition? Research results in acoustic modeling are typically evaluated based on performance on a single dataset. While the research community has coalesced around various benchmarks, we set out to understand generalization performance in acoustic modeling across datasets - in particular, if models trained on a single dataset transfer to other (possibly out-of-domain) datasets. We show that, in general, reverberative and additive noise augmentation improves generalization performance across domains. Further, we demonstrate that when a large enough set of benchmarks is used, average word error rate (WER) performance over them provides a good proxy for performance on real-world noisy data. Finally, we show that training a single acoustic model on the most widely-used datasets - combined - reaches competitive performance on both research and real-world benchmarks.
@article{arxiv.2010.11745,
title = {Rethinking Evaluation in ASR: Are Our Models Robust Enough?},
author = {Tatiana Likhomanenko and Qiantong Xu and Vineel Pratap and Paden Tomasello and Jacob Kahn and Gilad Avidov and Ronan Collobert and Gabriel Synnaeve},
journal= {arXiv preprint arXiv:2010.11745},
year = {2021}
}