This is the implementation of the paper and 12th place solution of Freesound General-Purpose Audio Tagging Challenge. Dataset can be downloaded at https://www.kaggle.com/c/freesound-audio-tagging/data and it should be extracted in the data folder. Model training is specified in pipeline.py and one can experiment with different inputs and available CNNs there. Stacking and model evaluation are given in ensembling.py and evaluation.py respectively. The paper itself is available at DCASEpaper.pdf.