In this work, a methodology to detect tool wear states during a milling operation through sound classification is presented. The sound was recorded during milling operations for endmills with different wear states (brand new, wear out and chipped); the wear states were determined by measuring their mass before each cutting operation. After machining, a transfer learning task was implemented for custom classification of the sound. The above, by using the VGG16 deep neural network architecture. The sound data, was represented as spectrogram images for the classification model training. Four different metrics, were used to measure the model performance, showing 97.5\% in the worst result. In addition, the results showed that sound data have enough information to train classification models for cutting tools wearing. Finally, the method presented in this work can be used for the development of monitoring tools for the support of machining workshops; thus increase the efficiency of their tools, raw materials and machining time.