We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.SD

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Electrical Engineering and Systems Science > Audio and Speech Processing

Title: One-shot Voice Conversion For Style Transfer Based On Speaker Adaptation

Abstract: One-shot style transfer is a challenging task, since training on one utterance makes model extremely easy to over-fit to training data and causes low speaker similarity and lack of expressiveness. In this paper, we build on the recognition-synthesis framework and propose a one-shot voice conversion approach for style transfer based on speaker adaptation. First, a speaker normalization module is adopted to remove speaker-related information in bottleneck features extracted by ASR. Second, we adopt weight regularization in the adaptation process to prevent over-fitting caused by using only one utterance from target speaker as training data. Finally, to comprehensively decouple the speech factors, i.e., content, speaker, style, and transfer source style to the target, a prosody module is used to extract prosody representation. Experiments show that our approach is superior to the state-of-the-art one-shot VC systems in terms of style and speaker similarity; additionally, our approach also maintains good speech quality.
Comments: Accepted by ICASSP 2022
Subjects: Audio and Speech Processing (eess.AS); Sound (cs.SD)
Cite as: arXiv:2111.12277 [eess.AS]
  (or arXiv:2111.12277v2 [eess.AS] for this version)

Submission history

From: Zhichao Wang [view email]
[v1] Wed, 24 Nov 2021 05:30:47 GMT (6196kb,D)
[v2] Mon, 21 Feb 2022 09:36:54 GMT (6197kb,D)

Link back to: arXiv, form interface, contact.