Speech recognition is a process to recognize recorded speech using computer. This research includes data processing using digital low-pass Butterworth filter, fast Fourier transform and artificial neural network using back-propagation algorithm. In this research, computer can only recognize two words. Those words are “DOOR” and “LIGHT”. Before being recorded, at anytime, undesired sound may occur and affect the spectrum of the desired speech. Separating the desired and undesired sound can be done by using digital low-pass Butterworth filter considering the undesired sound is high frequency. After filtering, the speech can be recognized by processing the sound file into a value that can be calculated by computer. This process is done by using a method known as Fast Fourier Transform. Artificial neural network is used on computer to learn the given speech so that it could recognize it. The purpose of this experiment is to recognize the word “DOOR” dan LIGHT” automatically without any keypress from user. Experiments have been done to analyze the performance of this system. From the experiments, in the artificial neural network, if the system uses 1000 epoch, 32 nodes for hidden layer and 256 normalized data, it can recognize 100% of the trained speech. Testing with untrained speech result is 70% for the word “DOOR” and 60% for the word “LIGHT”.