learn.ifs.name

wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations

Baevski et al., 2020, wav2vec 2.0 (Advances in Neural Information Processing Systems 33 (NeurIPS 2020))

Subscriber deck — free during early access

2 cards in “5.1 Low-Resource Labeled Data Evaluation” · whole deck

5.1 Low-Resource Labeled Data Evaluation

  1. How well does wav2vec 2.0 do with very little labelled data?
    LARGE pre-trained on LV-60k: with 10 minutes (48 recordings of ~12.5 s) 4.8/8.2 test WER with a Transformer LM; with 1 h 2.9/5.8; with 10 h 2.6/4.9; with 100 h 2.0/4.0, beating iterative self-training, which needs repeated labelling, filtering and retraining.
    • Baevski et al., 2020, wav2vec 2.0 (Advances in Neural Information Processing Systems 33 (NeurIPS 2020)), §5.1
  2. What did scaling model size and unlabelled data do in wav2vec 2.0?
    Both helped: LARGE beat BASE on all setups, most on the noisy test-other, and pre-training on 53k h (LV-60k) instead of 960 h gave large further gains.
    • Baevski et al., 2020, wav2vec 2.0 (Advances in Neural Information Processing Systems 33 (NeurIPS 2020)), §5.1

Source