This is the authors' abstract. We don't add key points for this paper.
Contrastive Predictive Coding (CPC), based on predicting future segments of based on past segments is emerging as a powerful algorithm for learning of speech signal. However, it still under-performs methods on unsupervised evaluation benchmarks. Here, we introduce, a time-domain data augmentation library and find that applying in the past is generally more efficient and yields better than other methods. We find that a combination of pitch, additive noise and reverberation substantially increase the of CPC (relative improvement of 18-22%), beating the reference-light results with 600 times less data. Using an out-of-domain dataset,-domain data augmentation can push CPC to be on par with the state of the on the Zero Speech Benchmark 2017. We also show that time-domain data consistently improves downstream limited-supervision phoneme tasks by a factor of 12-15% relative.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Kharitonov et al. (2020) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: