Arias generation method based on neural-symbolic dual driving and programmed logic mapping
By constructing an acoustic fingerprint mapping library and a stylized logical mapping loss function, the problems of genre feature reproduction and accompaniment misalignment in the generation of small-sample opera music in existing models are solved, and efficient opera music generation is achieved.
CN122392464APending Publication Date: 2026-07-14ZHEJIANG UNIV OF TECH
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHEJIANG UNIV OF TECH
- Filing Date
- 2026-03-30
- Publication Date
- 2026-07-14
Smart Images
Figure CN122392464A_ABST
Abstract
The application discloses a singing aria generation method based on neural-symbolic dual driving and programmed logic mapping, selects classic singing sections of various schools of opera to establish a reference audio library, extracts acoustic characteristics of each singing section, statistically analyzes style constraint features in the acoustic characteristics, generates a school and an acoustic fingerprint confidence interval mapping table, analyzes a natural language description requirement input by a user, generates a structured prompt word containing a style constraint feature confidence interval, retrieves a best matching singing section from the reference audio library, and finally generates a required singing aria according to the structured prompt word and acoustic characteristics of the best matching singing section.
Need to check novelty before this filing date? Find Prior Art