wav2vec-C: A Self-Supervised Model for Speech Representation Learning

• DOI: 10.21437/interspeech.2021-717
• Publication Date: 2021-08-27T05:59:39Z
• Authors: Samik Sadhu; Di He; Che-Wei Huang; Sri Harish Mallidi; Minhua Wu; Ariya Rastrow; Andreas Stolcke; Jasha Droppo; Roland Maas
• Keywords: Representation

Wav2vec-C introduces a novel representation learning technique combining elements from wav2vec 2.0 and VQ-VAE.Our model learns to reproduce quantized representations partially masked speech encoding using contrastive loss in way similar 2.0.However, the quantization process is regularized by an additional consistency network that reconstruct input features VQ-VAE model.The proposed self-supervised trained on 10k hours of unlabeled data subsequently used as encoder RNN-T ASR fine-tuned with 1k labeled data.This work one very few studies selfsupervised tasks large volume real far-field data.The wav2vec-C encoded achieve, average, twice error reduction over baseline higher codebook utilization comparison 2.0.