Can speed perturbation plus SpecAugment be outperformed by novel combinations of speech data augmentations for ASR? A low-resource evaluation

Data augmentation is crucial for automatic speech recognition, especially in low-resource settings. This paper introduces a time-domain augmentation method, FadeOutIn, designed to better capture the dynamics of spontaneous speech and remain compatible with classical techniques. Experimental results show that combining FadeOutIn with SpecAugment yields a relative improvement of 4.24% on the out-of-domain test set, demonstrating its compatibility with classical augmentation techniques. Furthermore, integrating FadeOutIn with SpecAugment and speed perturbation (the classic combination) achieves the lowest error rate among all tested configurations. Meanwhile, this paper also investigates the efficacy of various data augmentation strategies for Hungarian automatic speech recognition tasks, offering new insights into the use of single versus combined augmentation strategies in the same task.