The study investigates the use of cough characteristics to distinguish between tuberculosis (TB) and non-TB coughs. A dataset of 33,000 passive coughs and 1,600 forced coughs was collected from 149 subjects with pulmonary TB and 46 controls with other respiratory illnesses in Nairobi, Kenya. A ResNet18-based cough classifier was trained using images of passive cough scalograms, achieving a fivefold cross-validation sensitivity of 0.70 (±0.11 SD). The smartphone-based model performed better in subjects with higher bacterial load (ROC-AUC: 0.87 [95% CI: 0.87 to 0.88], P < 0.001) or lung cavities (ROC-AUC: 0.89 [95% CI: 0.88 to 0.89], P < 0.001). The results suggest that passive cough features can distinguish TB from non-TB subjects and are associated with bacterial burden and disease severity. The study highlights the potential of cough classifiers as a non-invasive tool for TB screening, particularly using smartphone recordings.The study investigates the use of cough characteristics to distinguish between tuberculosis (TB) and non-TB coughs. A dataset of 33,000 passive coughs and 1,600 forced coughs was collected from 149 subjects with pulmonary TB and 46 controls with other respiratory illnesses in Nairobi, Kenya. A ResNet18-based cough classifier was trained using images of passive cough scalograms, achieving a fivefold cross-validation sensitivity of 0.70 (±0.11 SD). The smartphone-based model performed better in subjects with higher bacterial load (ROC-AUC: 0.87 [95% CI: 0.87 to 0.88], P < 0.001) or lung cavities (ROC-AUC: 0.89 [95% CI: 0.88 to 0.89], P < 0.001). The results suggest that passive cough features can distinguish TB from non-TB subjects and are associated with bacterial burden and disease severity. The study highlights the potential of cough classifiers as a non-invasive tool for TB screening, particularly using smartphone recordings.