Unified speech representation learning

US12217745B2 · kind B2 · utility

Assignee

• MICROSOFT TECHNOLOGY LICENSING, LLC · US

Inventors

• Yao Qian · Dublin, US
• Yu WU · Hoherlehmer Straße, CN
• Kenichi Kumatani · Sammamish, US
• Shujie Liu · Cupertino, US
• Furu WEI · Redmond, US
• Nanshan Zeng · Bellevue, US
• Xuedong Huang · Bellevue, US
• Chengyi Wang · Beijing, CN

Key dates

• Filing date: Jul 3, 2023
• Grant date: Feb 4, 2025
• Expiry date: Jul 3, 2043

Classification

• Technology area (CPC G): Physics
• CPC primary: G10L2015/025
• WIPO field: Computer technology
• WIPO sector: Electrical engineering

Abstract

A system obtains a first training data set comprising labeled speech data or both labeled and unlabeled data corresponding to a high-resource data set as well as latent speech representations based on the first training data set. The system trains a machine learning model on the first training data set to learn phonetically aware speech representations corresponding to the first training data set. The system applies the latent speech representations to a transformer context network to generate contextual representations. The system aligns each of the contextual representations with a phoneme label to generate phonetically-aware contextual representations. The system causes a refinement engine to further refine the machine learning model.