A typical electric powered wheelchair is normally buttons or joystick operated which requires some degrees of hand movements. However, for severely physically disabled patients such as patients with high level spinal injury, their hand motions can be restricted. One of the alternative wheelchair control methods is to use speech control instead of joystick control. A speech control system would allow the user to operate the wheelchair with speech commands instead of hand movements. In order to perform speech recognition on the given commands, a front-end process on speech signals is implemented. Some of the major components include data acquisition, feature extraction, and data quantization. An endpoint detection algorithm based on energy analysis is used to isolate individual words. Linear predictive coding (LPC) and cepstral analysis are chosen to characterize speech signals. For data compression and classification, fast vector quantization (VQ) codebook design and search algorithms are implemented based on partial distortion theorem. For each isolated words, this front-end process would provide a speech recognition system with a sequence of indexes which is a compressed representation of the characteristics of the original speech. The basic algorithms, hardware and software design, the results are presented.
Tsou, Ling, "Manual Wheelchair Automator: Design of Front-end Process for a Speech Recognition System" (2009). EE 4BI6 Electrical Engineering Biomedical Capstones. Paper 16.