Multimodal depression risk monitoring method based on acoustic rhythm anomaly evolution embedding
By extracting stable rhythm features and removing non-pathological noise, and combining semantic anchoring and multi-scale convolution techniques, the expression weights of rhythm abnormal evolution data are dynamically adjusted, solving the problem of inaccurate rhythm feature extraction in existing technologies and achieving higher accuracy in depression identification and diagnosis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIHANG UNIV
- Filing Date
- 2026-05-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multimodal depression detection methods neglect acoustic rhythm information, fail to effectively isolate non-pathological noise, ignore the alignment relationship between speech rhythm and text content, resulting in inaccurate rhythm feature extraction, and fail to dynamically adjust fusion weights during the feature fusion stage.
By extracting stable rhythm features and removing non-pathological acoustic interference, semantic anchoring technology and soft dynamic time warping framework are used for cross-modal monotonic time alignment. Multi-scale convolution and sliding window techniques are combined to extract pathological speech rate evolution features. A multi-modal adaptive fusion architecture is constructed to dynamically adjust the expression weights of rhythm abnormal evolution data, and finally, depression risk stratification prediction is performed.
It significantly improves the accuracy and clinical robustness of depression identification, enhances the accuracy of pathological rhythm feature extraction, strengthens the ability to perceive the dynamic evolution of speech rhythm in users with depression, and optimizes auxiliary diagnostic results.
Smart Images
Figure CN122417093A_ABST