With the development of interface technologies in smart devices, voice assistants quickly gained popularity. These assistants are designed to use voice commands to provide a more convenient interaction with people. In this regard, one of the methods for implementing a voice assistant based on neural networks is proposed. Methods for implementing the main stages of creating a voice assistant have been studied. The article presents the results of testing a model based on a convolutional neural network. The following words were chosen as speech commands: yes, no, up, down, right, left, go, stop. This model classifies 8 speech commands with an accuracy of 86.63%. The neural network model best classified commands: yes, up, down, right, left, stop. The command to go is 66.67% accurate, and no is 76.4%, this is due to the similar sounding of the words down, go, no.