ADFF-Net: An Attention-Based Dual-Stream Feature Fusion Network for Respiratory Sound Classification

Deep learning–based respiratory sound classification has emerged as a promising non-invasive and cost-effective approach to assist clinical diagnosis. However, existing methods often face challenges, such as sub-optimal feature representation and limited model expressiveness. To address these issues, we propose an Attention-based Dual-stream Feature Fusion Network (ADFF-Net). Built upon the Audio Spectrogram Transformer, ADFF-Net takes Mel-filter bank and Mel-spectrogram features as dual-stream inputs, while an attention-based fusion module with skip connections is introduced to emphasize pathological spectral regions and preserve multi-scale time–frequency information. Extensive experiments on the ICBHI2017 database with the official train–test split show that ADFF-Net achieves superior performance in the four-class classification task, outperforming traditional fusion strategies and achieving results comparable to state-of-the-art approaches, with specificity of 81.39%, sensitivity of 42.91%, and overall accuracy of 62.14%. These findings highlight the effectiveness of dual-stream acoustic feature fusion and demonstrate the potential of ADFF-Net for clinical decision support in respiratory disease diagnosis.