Multi domain features for speaker independent word recognition

Currently, speech recognition technology is in a mature condition, however, compared with other applied technologies, speech recognition technology still has limited daily live applications. One of the reasons is that the lack of our knowledge regarding speech features. We performed word recognition experiments using features that are extracted by several different methods. We called them multi domain features. As shown in the results of recognition experiments, it is interesting that the multi domain parameters which represent phonemic features are very effective for word recognition and operate to promote the goodness of the recognition in cooperation with the other conditions like a distance measure.