Refer to this github repository address: https://github.com/fschmid56/EfficientAT. The main code idea was applied to this specific project ----- audio classification.
Training with tensorflow preprocessing and pytorch models; In subsequent inference, tensorflow preprocessing +onnx inference is used to achieve the goal of inference on the tensorflow architecture（without any pytorch code）.
CSDN记录：https://blog.csdn.net/qq_44799766/article/details/153833215