Incremental feedback learning methods for voice recognition based on DTW

A Dynamic Time Warping (DTW) based voice recognition approach and its template training problem are discussed in this paper. In order to achieve better recognition accuracies, we proposed two kinds of incremental feedback learning methods for DTW-based voice recognition, including the variable gain coefficient (VGC) based method and the time warping average (TWA) based method. Compared with the non-feedback recognition system, the proposed methods are easy to implement and more robust to noise. A number of comparison experiments are performed to demonstrate the effectiveness of the proposed method. It is also shown that the VGC-based method, whose implementation is easy, achieves significantly better performance than conventional batch template training method without feedback learning.