RAMoEA-QA: Hierarchical Specialization for Robust Respiratory Audio Question Answering
Researchers have introduced RAMoEA-QA, a hierarchical mixture-of-experts architecture designed to improve robustness in respiratory audio question answering. The system addresses heterogeneity in audio recordings from varying devices, environments, and protocols, as well as diverse query types. Unlike existing single-path models, RAMoEA-QA employs specialized pathways for different recording conditions and question formats. The model is evaluated on multiple datasets, demonstrating superior performance and robustness compared to baselines. This work advances conversational AI in healthcare, particularly for non-invasive respiratory monitoring.
Key facts
- RAMoEA-QA is a hierarchical mixture-of-experts architecture for respiratory audio question answering.
- It addresses heterogeneity in audio recordings across devices, environments, and protocols.
- The system uses specialized pathways for different recording conditions and query types.
- It outperforms single-path models in robustness and accuracy.
- Evaluated on multiple datasets.
- Published on arXiv with ID 2603.06542.
- Focuses on non-invasive respiratory monitoring via audio.
- Aims to improve clinical decision support in respiratory care.
Entities
Institutions
- arXiv