5th International Conference on Spoken Language Processing
Feature representations mediating between acoustic input and symbolic representation promise to reduce learning time needed for automatic speech signal segmentation. Experiments are reported that circumscribe simple acoustic inputs and appropriate feature sets for neural network training. Stable and compatible solutions for English and French were identified.
Bibliographic reference. Keller, Eric (1998): "Neural network motivation for segmental distribution", In ICSLP-1998, paper 0937.