一种语音处理方法、装置、介质机计算机设备
By processing speech signals using multi-scale convolutional neural networks, the problems of high hardware costs and limited noise reduction effects in existing technologies are solved, achieving efficient speech noise reduction, especially in improving dynamic noise.
CN115910089BActive Publication Date: 2026-07-17WUHAN DOUYU NETWORK TECHNOLOGY CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- WUHAN DOUYU NETWORK TECHNOLOGY CO LTD
- Filing Date
- 2021-08-20
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing technologies for speech noise reduction have high hardware costs and limited noise reduction effects, especially for dynamic noise.
Method used
Multi-scale convolutional neural networks are used for speech signal processing, including convolution, normalization, multi-scale feature fusion, superposition, and high-dimensional embedding vector processing. The receptive field is increased by multi-scale convolutional neural networks to improve speech enhancement performance.
Benefits of technology
Without increasing hardware costs, it significantly improves voice noise reduction and enhances the ability to handle dynamic noise.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115910089B_ABST
Abstract
本发明提供一种语音处理方法、装置、介质机计算机设备，方法包括：对初始带噪语音信号进行卷积并激活，获得第一带噪语音信号；对第一带噪语音信号进行归一化、卷积处理，获得第二带噪语音信号；利用多尺度卷积神经网络对第二带噪语音信号进行多尺度特征融合，获得第三带噪语音信号；对第一带噪语音信号及第三带噪语音信号进行叠加，获得第四带噪语音信号；对第四带噪语音信号进行处理，获得噪声高维嵌入向量及降噪高维嵌入向量；对噪声高维嵌入向量及降噪高维嵌入向量进行处理，获得分离噪声及分离语音；如此，对第二带噪语音信号进行多尺度特征融合，因此可增大卷积神经网络的感受野，进而提高卷积神经网络的语音增强性能，确保语音降噪性能。
Need to check novelty before this filing date? Find Prior Art