Abstract:
Embedded systems have become common in many devices or systems. In these systems. natural
communication between the systems and humans are preferred. This paper presents a hardware
implementation of the natural voice-based human machine interface for such embedded systems.
We propose the hardware framework with a design methodology that will work for different
applications. The proposed design is based on the fact that the communication between human and a
machine needs only a small subset of a language. Moreover. at any specific time, the machine only
looks for a smaller subset of words. Therefore, we groups N words that are needed to be recognized
into M subsets, where the maximum number of words per subset is L. As the number of words to be
recognized at each state is small, the chance of correct recognition is increased. We demonstrate the
design method using a wheelchair control application and smart home automation. The MFCC with
the dynamic time wrapping (DTW) is chosen as the feature extraction and classification of words,
respectively. The results show that the overall recognition rate of the proposed method is 88.69
percent, which gives 8.57 percent improvement by average over the conventional method.