Machine was trained on given 40k images and then model was built. Images were encoded - Attention model was applied and decoded. Predicted caption was converted to mp3 with the help of gtts.