#################################################### # Keywords Recognition via MFCC+DTW # # Author: Bill Xia # Blog: http://ibillxia.github.com # Email: ibillxia@gmail.com #################################################### This project is a simple keywords recognition system by DTW(Dynamic Time Warping). The MFCC feature extraction part is refered from SPro5.0, which is an open source project for speech feature extraction. You can regard it as a simple version of SPro5.0 for cepstral analysis. The DTW algorithm is very simple here and havn't been optimized. To use it, you should first get some speech for training and choose the best one as the model for each command. The criteria of choosing the best sentence is off to you. Here I choose the class center point(centroid) as the best one. Then you can use it in the real applications. Performance: for the three commands in my test session, only 2 errors in 39 test case, this means the overall error rate is only 5.128%. Copyright (C) 2013 Bill Xia (ibillxia@gmail.com)