Neural Network Control Interface of the Speaker Dependent Computer System "Deep Interactive Voice Assistant DIVA" to Help People with Speech Impairments

With the development of modern informational communication systems, voice control interface and speech recognition systems find application in various fields of activity. One application of such systems is for people with special needs who have speech impairments, and thus find using speech-dependent voice interfaces challenging. Our research team is developing a speaker dependent computer system "Deep Interactive Voice Assistant" (DIVA), which allows recognizing an arbitrary set of commands to control the computing system. The article presents the results of testing various artificial neural networks to train the machine to recognize vocal inputs. We examine such architectures as associative memory, multilayer perceptron and convolutional network. The research justifies the use of multilayer perceptron for the speaker dependent computer system DIVA as a training solution that demonstrated high results on a small selection. DIVA will be implemented in voice-user interface of such systems as "Smart House", mobile applications and IT-based assistive systems.