Speaker-independent vowel-sequence recognition based on mutual relationships between vowels

A vowel recognition technique that deals with input vowels as one sequence and recognizes them using binary relationships between data is investigated. Speaker-independent recognition experiments on monosyllabic vowels and vowels in words showed that recognition rates are 99.1% and 93.2%, respectively. These results are about 2% better than those obtained using a statistical matching method, for instance, a Bayes classifier, which discriminates each vowel individually. Furthermore, it is shown that the presented method works well even under the vowel-sequences which contain segmentation errors