一种基于时频自适应与协同注意力的声音事件分类方法
By employing time-frequency adaptive and collaborative attention mechanisms, a sound event classification model is constructed, which solves the problem of low classification accuracy of existing models in complex acoustic scenarios. This model improves classification accuracy and modeling ability without increasing the number of parameters, and enhances the feature extraction capability for complex acoustic scenarios.
Patent Information
- Application Number
- CN202610850070.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing sound event classification models are not accurate enough in complex acoustic scenarios, and it is difficult to improve global and local modeling capabilities without increasing the number of parameters. Furthermore, existing attention mechanisms are difficult to balance computational efficiency with long-distance dependency modeling capabilities.
Employing a time-frequency adaptive and collaborative attention mechanism, a sound event classification model is constructed using a time-frequency decoupled adaptive module (TFDA), a cross-scale collaborative attention module, a gating module, a multilayer perceptron (MLP), and a global average pooling (GAP). Combined with time-frequency separation dynamic convolution and a multi-scale expanded deep residual network, efficient feature extraction of complex acoustic patterns is achieved.
Without significantly increasing the number of parameters, it improves the accuracy of sound event classification and global and local modeling capabilities, and enhances the feature representation capabilities of complex acoustic scenes.
Smart Images
Figure CN122417082A_ABST