Revisiting Interpolation Augmentation for Speech-to-Text Generation
About
Speech-to-text (S2T) generation systems frequently face challenges in low-resource scenarios, primarily due to the lack of extensive labeled datasets. One emerging solution is constructing virtual training samples by interpolating inputs and labels, which has notably enhanced system generalization in other domains. Despite its potential, this technique's application in S2T tasks has remained under-explored. In this paper, we delve into the utility of interpolation augmentation, guided by several pivotal questions. Our findings reveal that employing an appropriate strategy in interpolation augmentation significantly enhances performance across diverse tasks, architectures, and data scales, offering a promising avenue for more robust S2T systems in resource-constrained settings.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Automatic Speech Recognition | LibriSpeech (test-other) | WER44.76 | 966 | |
| Automatic Speech Recognition | LibriSpeech (dev-other) | WER43.89 | 411 | |
| Automatic Speech Recognition | LibriSpeech 960h (test-other) | WER7.51 | 81 | |
| Automatic Speech Recognition | AISHELL-1 (test) | -- | 71 | |
| Automatic Speech Recognition | LibriSpeech 960h (test-clean) | WER0.0301 | 53 | |
| Automatic Speech Recognition | LibriSpeech 960h (dev-other) | WER7.61 | 50 | |
| Speech Translation | MuST-C EN-DE (test-COMMON) | BLEU27.5 | 41 | |
| Automatic Speech Recognition | LibriSpeech 960h clean (dev) | WER2.91 | 25 | |
| Speech Translation | MuST-C en-de (dev) | BLEU26.92 | 14 | |
| Automatic Speech Recognition | LibriSpeech clean 10h (dev) | Word Error Rate0.2834 | 2 |