Speaker-specific adaptation of a machine-learned speech synthesis model for voiceless speakers
By using text-to-speech models to generate synthetic speech for voiceless speakers and capturing corresponding measurement data streams, the method addresses the challenge of speaker-specific adaptation, enhancing speech synthesis accuracy for voiceless individuals.
WO2026146123A1PCT designated stage Publication Date: 2026-07-09ALTAVO GMBH
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- ALTAVO GMBH
- Filing Date
- 2025-12-29
- Publication Date
- 2026-07-09
AI Technical Summary
Technical Problem
Existing machine-learned speech synthesis models struggle to achieve accurate speaker-specific adaptation, particularly for voiceless speakers, due to the lack of recorded audio signals and anatomical and technical variations.
Method used
A method for generating speaker-specific training samples using text-to-speech models to create synthetic speech utterances, which are then used to capture measurement data streams for voiceless speakers, allowing iterative refinement of the model to adapt to individual speakers.
Benefits of technology
Enables accurate speaker-specific adaptation of machine-learned speech synthesis models for voiceless speakers, accounting for anatomical and technical changes, and improving speech synthesis accuracy.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure EP2025089067_09072026_PF_FP_ABST
Abstract
Techniques are described that allow the generation of speaker-specific training samples in a training data set for training a machine-learned speech synthesis model also for voiceless speakers. A speaker-specific training of the machine-learned speech synthesis model is made possible in this way. The machine-learned speech synthesis model can regularly be trained further, for example if the anatomy of a particular speaker and / or the hardware of the measurement modality used changes. As a result, it is possible to provide speech synthesis with increased accuracy.
Need to check novelty before this filing date? Find Prior Art