This project about sign language conversation interpreter system. It is made for speak disable people. Here a man who don't know the sign language can make conversation with a speak disable people. In this project we used efficient methods to convert sign language into text. We used our own dataset and collected dataset of Bangla Sign Languages using hand gestures. Inputs will be taken by webcam and recognized by neural network system. System will show text output by using gestures. We have created a CNN which is similar model of the MNIST classifying model. It can use both Tensorflow and Keras. This model was trained using Keras by video streaming. In a video stream sign language, interpreter can turn it fast into text. We used histogram back projection method to recognize people's hand as object. For better result we calculated the grey level of hands.