Recognition of Danish phonemes using an artificial neural network
S. Danielson · 1990
Describes the utilization of an artificial neural network for the recognition of Danish phonemes and presents the results obtained. The artificial neural network implementation is based on Kohonen's (1988) self-organizing feature maps, which have shown superior results in recognition of Finnish and Japanese phonemes. The aim of the work is to design a robust front-end processing system which will be integrated into a continuous speech recognition system. The results obtained show that a self-organizing feature map consisting of 225 nodes is suitable for classification of fourteen vowel and ten consonant phonemes embedded in naturally spoken Danish sentences, showing a recognition rate of 68% on the frame level and 74% on the segment level