Voice-activated powered wheelchair for severely disabled persons

In this paper, we consider that non-voice inputs such as coughing and breathing and spark-like mechanical noise should be rejected and that the system must recognize speech commands affected by disability, including specific pronunciation speed and frequency, to achieve a system that can operate powered wheelchairs safely using a severely disabled person's voice in a real environment. We therefore propose inarticulate speech recognition using a multi-template dictionary to cope with the pronunciation variation and perform non-voice rejection using both YIN-based fundamental frequency extraction and reliability in preprocessing.