The study dealt with the development and validation of a screening model for the detection of dysphagia in the elderly using the acoustic properties of voice, cough and swallowing. Audio data were obtained from 419 elderly people from nursing homes in the Beijing area and used to select key features using LASSO regression. The models were built using multiple machine learning methods including Logistic Regression, Random Forest, SVM and XGBoost. The XGBoost model achieved the best performance with an AUC of 0.86 in internal validation and an AUC of 0.71 in external validation on 216 seniors from the Shijiazhuang area. The model demonstrated good ability to distinguish dysphagia risk, calibration and clinical applicability. This tool can help healthcare professionals identify elderly patients at high risk of dysphagia for early intervention and improved clinical outcomes. The results confirm the potential of automated dysphagia screening based on the analysis of sound signals in the elderly.