Online monitoring of earthquake in a seismological network involves detection and picking phases, association and location tasks. In recent years, accuracy of each individual step increased by the use of machine learning. However, an end-to-end procedure of all of these steps are also possible using neural networks. In such procedure, seismic waveforms are imported to a neural network and both event and location are detected without implementing the intermediate picking and location steps. In this study, the data of one year (2018) record of 27 stations in the Iranian seismological network are used. Earthquakes events and locations are extracted using available catalogues as training dataset. To reduce the dimension of the training data and extracting more effective features from waveforms, popular filters in speech detection are used. First network is so trained that could detect earthquake events from the patterns of input data by an accuracy of more than 98%. Second network uses this event as the input data and could estimate earthquake location with an accuracy more than 99% and a mean absolute error of 25 km in each direction.