← All works
Research paper

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

At a glance

Learned from unlabeled audio to reduce the amount of transcribed speech needed for recognition.

In one paragraph

Introduced a speech-recognition method that learns from unlabeled recordings before being trained on a small amount of transcribed speech. It reduced the need for costly human-written transcripts.

Authors 4

  1. Alexei Baevski
  2. Henry Zhou
  3. Abdelrahman Mohamed
  4. Michael Auli

Organizations

Credited for this work at publication.