This research has developed a simple and efficient algorithm for Bangla speech recognition using an artificial neural network. This supervised learning model consists of four major phases: collecting audio samples, conversion into frequency domain & filtering, normalization, and applying the back propagation algorithm to train the artificial neural network. Proper filtering has granted higher noise immunity and has decreased the number of elements in the input layer of the neural network. Thus, the overall processing speed is increased. The audio samples for this research have been collected through a digital microphone, and the complete model is illustrated using GNU Octave. Although this research focuses on the Bangla language, this model can be applied to any other language.