This paper describes the utilization of multilayer perceptrons and radial basis function networks for the syllable segmentation of continuous speech by detecting the syllable nuclei. The artificial neural nets were trained to indicate the occurrence of vowels or diphthongs by means of the backpropagation algorithm and a non-iterative matrix-inversion method. In different experiments the syllable nuclei were correctly marked with more than 93 % and less than 2 % insertions.
Keywords: Syllable segmentation, neural networks, multilayer perceptrons, radial basis functions