In the surface anomaly detection of industrial products, due to the scarcity of abnormal samples, unsupervised methods have become mainstream approaches, and the detection ability is enhanced by introducing anomaly synthesis strategy. Currently, the method of adding Gaussian noise to normal features to simulate abnormal features has achieved excellent results, but the control of noise scale has become a difficult problem. Therefore, we propose a noise scale controllable anomaly feature synthesis strategy (NSCAS). The strategy consists of three parts: noise scale control, abnormal mask generation and abnormal feature synthesis. The NSCAS model achieves advanced detection results on MVTec-AD (I-AUROC of 99.6%), VisA (P-AUROC of 98.4%) datasets. Its effectiveness and generalization ability have been further verified in the self-built speaker shell anomaly dataset (SS-AD).