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

Download:

Current browse context:

eess.AS

Change to browse by:

References & Citations

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: Speech Quality Assessment through MOS using Non-Matching References

Abstract: Human judgments obtained through Mean Opinion Scores (MOS) are the most reliable way to assess the quality of speech signals. However, several recent attempts to automatically estimate MOS using deep learning approaches lack robustness and generalization capabilities, limiting their use in real-world applications. In this work, we present a novel framework, NORESQA-MOS, for estimating the MOS of a speech signal. Unlike prior works, our approach uses non-matching references as a form of conditioning to ground the MOS estimation by neural networks. We show that NORESQA-MOS provides better generalization and more robust MOS estimation than previous state-of-the-art methods such as DNSMOS and NISQA, even though we use a smaller training set. Moreover, we also show that our generic framework can be combined with other learning methods such as self-supervised learning and can further supplement the benefits from these methods.
Comments: To Appear, Interspeech 2022
Subjects: Audio and Speech Processing (eess.AS); Sound (cs.SD)
Cite as: arXiv:2206.12285 [eess.AS]
  (or arXiv:2206.12285v1 [eess.AS] for this version)

Submission history

From: Pranay Manocha [view email]
[v1] Fri, 24 Jun 2022 13:34:53 GMT (2803kb,D)

Link back to: arXiv, form interface, contact.