METHOD FOR TRAINING A SPEECH SYNTHESIS MODEL, SPEECH SYNTHESIS METHOD AND ASSOCIATED DEVICES
Patent Information
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- TENCENT TECHNOLOGY (SHENZHEN) CO LTD
- Filing Date
- 2023-07-24
- Publication Date
- 2026-06-10
AI Technical Summary
Existing speech synthesis models require significant computing resources due to their large parameter count, making deployment in devices with low computing power challenging, particularly in scenarios like terminals.
A speech synthesis model is trained using a first submodel to predict phonetic posteriorgrams and a second submodel to generate speech through inverse Fourier transform, reducing the need for spectral features and upsampling, thereby minimizing model parameters and computational complexity.
The approach allows for the generation of speech that matches a target user's timbre with reduced computing resources, enabling deployment on devices with limited capabilities by utilizing fewer model parameters and inverse Fourier transform.