Fuzzification of formant trajectories for classification of CV utterances using neural network models

The paper reports the results of a study which shows that fuzzification of input and output data improves the recognition accuracy of CV segments. In particular, fuzzification of input data taking into account the fact that the formant data is for a sequence of frames, improves the recognition of CV segments significantly. In the studies only a simple method was used to implement the dependence of fuzziness on the sequence. But a more sophisticated data dependent approach for determining the fuzzy membership values for data both along frequency and along time may improve the recognition performance still further.