A real-time speech denoising method and system based on deep neural networks
By fusing time-domain and frequency-domain information of speech signals through generative adversarial networks, a generator and a discriminator are constructed. This solves the problems of high computational cost and low real-time performance of deep learning algorithms in non-stationary noise processing, and achieves efficient real-time speech denoising effect.
CN121148407BActive Publication Date: 2026-05-26COMMUNICATION UNIVERSITY OF CHINA
View PDF 2 Cites 0 Cited by
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- COMMUNICATION UNIVERSITY OF CHINA
- Filing Date
- 2025-10-28
- Publication Date
- 2026-05-26
AI Technical Summary
Technical Problem
Existing deep learning algorithms suffer from high computational cost and low real-time performance when dealing with non-stationary noise, and their denoising performance is also low for noise types that are not covered.
Method used
By employing a generative adversarial network, which integrates time-domain and frequency-domain information of speech signals, a generator and a discriminator are constructed. The model is then optimized through adversarial training and a joint loss function to achieve real-time speech denoising.
Benefits of technology
The generated speech has thorough background noise suppression, excellent noise reduction performance, good robustness and generalization ability, and meets the requirements of real-time processing.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN121148407B_ABST
Abstract
This invention discloses a real-time speech denoising method and system based on deep neural networks, applied in the field of speech denoising technology. The method includes the following steps: acquiring clean speech samples; adding random noise to the clean speech samples to obtain noisy speech samples; preprocessing the speech samples to obtain a training dataset; constructing a generative adversarial neural network for real-time speech denoising, including a generator and a discriminator; iteratively training the generator and discriminator based on the training dataset and evaluating the network performance; deploying the generator, acquiring noisy speech signals in real time and preprocessing them; inputting the preprocessed noisy speech into the generator to obtain a denoised speech signal. This invention meets the real-time processing requirements while ensuring speech denoising performance.
Need to check novelty before this filing date? Find Prior Art