Voice Controlled Wheelchair for Disabled Patients based on CNN and LSTM

Abstract

Disabled patients with reduced mobility due to health problems such as disability, injury, paralysis, or other factors will experience difficulty in movement. They need tools that can help them, in which the most widely used is a wheelchair. The main objective of this research is to control wheelchair motion with voice commands. There are five commands for wheelchair control: forward, backward, right, left, and stop. Voice data is obtained from recording several subjects using Sound Recorder Pro and Sox Sound Exchange. The voice commands for wheelchair navigation were identified using Convolutional Neural Network (CNN) and Long Short- Term Memory (LSTM) combination embedded in Raspberry Pi 3. Voice data is first converted to spectrogram images before being fed into CNN using Mel-Frequency Cepstrum Coefficients (MFCC). This system can be controlled by simple voice commands given by the user. This method is proven to be useful in speech recognition with an accuracy level using CNN-LSTM above 97.80 %. Preliminary experimental results indicate that voice commands in wheelchairs using the CNN-LSTM can be recognized well.