Voice high frequency component reconstruction method, device and computer equipment
By combining traditional algorithms and neural networks, and utilizing the characteristics of consonant zero-crossing rate to introduce white noise and half-wave rectification, fine adjustment of high-frequency components is achieved, solving the problems of distortion in traditional algorithms and large number of neural network parameters, thus improving the quality and stability of speech reconstruction.
Patent Information
- Application Number
- CN202610358482.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-23
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, traditional algorithms can only restore the high-frequency components of vowels, resulting in distortion, and cannot reconstruct consonants, while neural network solutions have a large number of parameters and are difficult to implement on the edge.
A hybrid approach is adopted, which utilizes the consonant zero-crossing rate characteristics of traditional algorithms to introduce white noise and half-wave rectification, and combines neural networks to finely adjust high-frequency components, thereby constructing a model with fewer parameters.
It improves the quality and stability of high-frequency speech component reconstruction, has fewer model parameters, is easy to implement on the edge, has low latency, and meets the requirements of real-time signal processing.
Smart Images
Figure CN122417063A_ABST
Abstract
Citation Information
Patent Citations
Processing of audio signals during high frequency reconstruction
CN103155033A
Integration of high-frequency reconstruction techniques that reduce post-processing latency.
MOJ009722C
Method and apparatus for generating excitation signal and regenerating signal in bandwidth extension
WO2009076871A1