一种单通道语音回声消除方法和装置
By using GRU to construct full-band and sub-band models in audio and video communication systems, the problems of complexity of deep neural network models and lack of consideration of frequency correlation are solved, achieving low-consumption and high-efficiency echo cancellation effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- G NET INTEGRATED SERVICE
- Filing Date
- 2022-11-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing echo cancellation methods based on deep neural networks suffer from problems such as model complexity, large model size, and failure to consider inter-frequency correlation in remote audio and video conferencing systems, resulting in poor echo cancellation performance and near-end speech impairment.
A full-band and sub-band model based on gated cyclic units (GRU) is adopted. By splicing the near-end and far-end frequency domain signal features and dividing them into sub-bands, and combining the GRU layer and fully connected layer for optimization calculation, speech echo cancellation is achieved.
It achieves efficient echo suppression with low performance consumption and small model size, and is suitable for real-time audio and video communication systems, with good echo cancellation effect.
Smart Images
Figure CN115938379B_ABST