Features for QbE-STD: wav2vec 2.0 LibriSpeech 960 Transformer Layer 11 (Archive III: top performing features)

Nay San, Martijn BarteldsZenodo 2021 · DOI · OpenAlex

Abstract

Best performing features extracted from audio using Transformer layer 11 of the wav2vec 2.0 LibriSpeech 960h on all 10 datasets (eng-mav, gbb-lg, gbb-pd, gos-kdl, gup-wat, mwf-jm, pjt-sw01, wbp-jk, wrl-mb, wrm-pd).
BibTeX
@inproceedings{san2021features,
  author = {Nay San and Martijn Bartelds},
  title = {{Features for QbE-STD: wav2vec 2.0 LibriSpeech 960 Transformer Layer 11 (Archive III: top performing features)}},
  booktitle = {Zenodo},
  year = {2021},
  doi = {10.5281/zenodo.4638384},
}

Loading related content...

Figure