Abstract
This paper suggest an original approach, based on event-driven processing, for time-domain spoken English letter features extraction and classification. The idea is founded on smartly combining the event-driven signal acquisition and segmentation along with local features extraction and voting based classification for realizing an efficient and high precision solution. The incoming spoken letter is digitized with an event-driven A/D converter (EDADC). An activity selection mechanism is employed to efficiently segment the EDADC output. Later on, features of these segments are mined by performing the time-domain analysis. The recognition is done with a specifically developed voting based classifier. The classification algorithm is described. The system functionality is tested for a case study and results are presented. A 9.8folds reduction in accumulated count of samples is achieved by the devised approach as compared to the traditional counterparts. It aptitudes a significant processing gain and efficiency increase in terms of utilization of power of the suggested approach in contrast to the counterparts. The proposed system attains an average subject dependent recognition accuracy of 92.2%.It demonstrates the potential of using the suggested solution for the realization of computationally efficient automatic speech recognition applications.