arXiv | Conference | BibTeX
To train or evalutate the audio diffusion model, please see Diffwave/
folder.
To train or evalutate the respiratory sound classification task, please see Classification/
folder.
Ubuntu xx.xx
Python 3.8.xx
pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2
and
pip install -r requirements.txt
Download the ICBHI files and unzip it. All details is described in the paper w/ code
wget https://bhichallenge.med.auth.gr/sites/default/files/ICBHI_final_database/ICBHI_final_database.zip
or
wget --no-check-certificate https://bhichallenge.med.auth.gr/sites/default/files/ICBHI_final_database/ICBHI_final_database.zip
To directly use our pre-processed ICBHI dataset for reproducibility, see in Diffwave/
and Classification/
folder.
The database consists of a total of 5.5 hours of recordings containing 6898 respiratory cycles, of which 1864 contain crackles, 886 contain wheezes, and 506 contain both crackles and wheezes, in 920 annotated audio samples from 126 subjects.
The downloaded data looks like [kaggle, paper w/ code]:
data/icbhi_dataset ├── metadata.txt │ ├── Patient number │ ├── Age │ ├── Sex │ ├── Adult BMI (kg/m2) │ ├── Adult Weight (kg) │ └── Child Height (cm) │ ├── official_split.txt │ ├── Patient number_Recording index_Chest location_Acqiosotopm mode_Recording equipment │ | ├── Chest location │ | | ├── Trachea (Tc),Anterior left (Al),Anterior right (Ar),Posterior left (Pl) │ | | └── Posterior right (Pr),Lateral left (Ll),Lateral right (Lr) │ | | │ | ├── Acquisition mode │ | | └── sequential/single channel (sc), simultaneous/multichannel (mc) │ | | │ | └── Recording equipment │ | ├── AKG C417L Microphone (AKGC417L), │ | ├── 3M Littmann Classic II SE Stethoscope (LittC2SE), │ | ├── 3M Litmmann 3200 Electronic Stethoscope (Litt3200), │ | └── WelchAllyn Meditron Master Elite Electronic Stethoscope (Meditron) │ | │ └── Train/Test │ ├── patient_diagnosis.txt │ ├── Patient number │ └── Diagnosis │ ├── COPD: Chronic Obstructive Pulmonary Disease │ ├── LRTI: Lower Respiratory Tract Infection │ └── URTI: Upper Respiratory Tract Infection │ └── patient_list_foldwise.txt
(Real Test Sample) 170_1b2_Tc_mc_AKGC417L_event_3_label_0.wav
(Real Test Sample) 176_2b3_Ar_mc_AKGC417L_event_2_label_1.wav
(Generated Test Sample) 176_2b3_Ar_mc_AKGC417L_event_2_label_1.wav
(Real Test Sample) 178_1b2_Lr_mc_AKGC417L_event_11_label_1.wav
(Generated Test Sample) 178_1b2_Lr_mc_AKGC417L_event_11_label_1.wav
(Real Test Sample) 216_1b1_Al_sc_Meditron_event_2_label_1.wav
(Generated Test Sample) 216_1b1_Al_sc_Meditron_event_2_label_1.wav
(Real Test Sample) 147_2b2_Pl_mc_AKGC417L_event_2_label_2.wav
(Generated Test Sample) 147_2b2_Pl_mc_AKGC417L_event_2_label_2.wav
(Real Test Sample) 218_1b1_Pr_sc_Meditron_event_2_label_2.wav
(Generated Test Sample) 218_1b1_Pr_sc_Meditron_event_2_label_2.wav
(Real Test Sample) 223_1b1_Pr_sc_Meditron_event_7_label_2.wav
(Generated Test Sample) 223_1b1_Pr_sc_Meditron_event_7_label_2.wav
(Real Test Sample) 151_2p3_Al_mc_AKGC417L_event_2_label_3.wav
(Generated Test Sample) 151_2p3_Al_mc_AKGC417L_event_2_label_3.wav
(Real Test Sample) 195_1b1_Ll_sc_Litt3200_event_3_label_3.wav
(Generated Test Sample) 195_1b1_Ll_sc_Litt3200_event_3_label_3.wav
(Real Test Sample) 218_1p1_Pl_sc_Litt3200_event_6_label_3.wav
(Generated Test Sample) 218_1p1_Pl_sc_Litt3200_event_6_label_3.wav
If you find this repo useful for your research, please consider citing our paper:
@article{kim2023adversarial,
title={Adversarial Fine-tuning using Generated Respiratory Sound to Address Class Imbalance},
author={Kim, June-Woo and Yoon, Chihyeon and Toikkanen, Miika and Bae, Sangmin and Jung, Ho-Young},
journal={arXiv preprint arXiv:2311.06480},
year={2023}
}
- June-Woo Kim: kaen2891@gmail.com