New PASQA model targets pitch-accent errors in speech quality assessment

Researchers have developed a new model called PASQA, designed to specifically assess speech quality by focusing on pitch-accent errors. Unlike existing models that predict overall naturalness, PASQA is trained on a custom Japanese dataset with controlled accent errors generated via speech synthesis. This approach allows PASQA to accurately rank speech by accent error severity and better align with human judgments on accent correctness. AI

IMPACT Introduces a specialized model for evaluating speech quality, potentially improving automated assessment systems.

RANK_REASON Academic paper detailing a new model for speech quality assessment. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Japanese

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →