Speech Command Based Intelligent Control of Multiple Home Devices for Physically Handicapped

In this study, a Convolutional Neural Network (CNN) model has been developed for speech command based intelligent control, where physically handicapped people can actively use any home devices they need. Firstly, a dataset containing 14 different Turkish speech commands, including datasets such as ’Aç (Open)’ and ’Kapa (Close)’, was created to train the CNN model in the study. Various methods for extracting the feature spectra of speech data have been proposed in the literature. In this study, features of speech data were extracted with Short-Term Fourier Transform-Chroma (STFTC), Spectral-Contrast, Melspectrogram and Mel Frequency Cepstral Coefficient (MFCC) models in the Librosa library. As a result, 14 different speech words were classified by the most successful MFCC based CNN model. With the MFCC based CNN model developed for speech classes, training and test processes were successfully carried out in the Kaggle Notebook environment. As a result, the control of the fan and LED night light system was successfully classified with an average accuracy of %97.42 Then, the most accurate MFCC based CNN speech recognition model software developed was loaded on the Raspberry Pi 3B model. Then, the model output created with the “Tensorflow Lite” model was transferred to the Raspberry Pi 3B model. Finally, the realized model was tested in the “Aç (Open)” and “Kapa (Close)” control of the real-time fan and LED night light system.