The thesis contains a detailed description of everything involved when developing an Android application that performs speech recognition with the help of deep neural networks. The focus lies mainly on developing a keyword spotting system that runs on a computationally constrained environment, in this case an Android application but the idea and technologies can be applied on other platforms as well, such as iOS or Raspberry Pi. The task of a keyword spotting system is to detect specific predefined keywords on a vocal sound. The thesis includes a thorough explanation of how audios are stored digitally, what kind of utilities does the Android platform provides for performing audio recordings, how are the audio inputs preprocessed before being fed into a neural network, how to build a neural network architecture that is capable of performing keyword spotting and how to prepare a data set of audio files that is used to train a neural network.