Remixit: continual self-training of speech enhancement models via bootstrapped remixing

HIGHLIGHTS

• who: Efthymios Tzinis and collaborators from the (UNIVERSITY) have published the article: RemixIT: Continual Self-Training of Speech Enhancement Models via Bootstrapped Remixing, in the Journal: (JOURNAL)
• what: The authors propose RemixIT which is based on several aforementioned state-of-the-art SSL strategies for pseudolabeling and continual training while also providing a novel technique for training speech enhancement models with OOD data. Since the authors are mostly interested in denoising, the authors focus on the speech estimates of the teacher and the student networks as with initial mixtures M and the bootstrapped mixtures M . . .
  

   

  

  If you want to have access to all the content you need to log in!

  Thanks :)

  Username or E-mail:

  If you don't have an account, you can create one here.