Phoneme dividing method using multilevel neural network

A phoneme dividing method using a multilevel neural network applied to a phoneme dividing apparatus having a voice input portion for outputting a user's digitally converted vocal sample, a preprocessor for extracting a characteristic vector suitable for phoneme division from the vocal sample input at the voice input portion, a multi-layer perceptron (MLP) phoneme dividing portion for finding and outputting the phoneme border using the characteristic vector of the preprocessor, and a phoneme border outputting portion for outputting position information regarding the phoneme border of the MLP phoneme dividing portion in the form of frame position, said method comprising the steps of:

• (a) sequentially segmenting and framing a voice with digitalized voice samples, extracting characteristic vectors by vocal frames, and extracting an inter-frame characteristic vector of the difference between nearby frames of the characteristic vector by frames, to thereby normalize the maximum and minimum of said inter-frame characteristic vectors;
• (b) initializing weights present between an input layer and a hidden layer and between the hidden layer and an output layer of said MLP, designating an output target data of said MLP, inputting said inter-frame characteristic vectors extracted from a currently analyzed frame to said MLP for learning, and storing and finishing information on the weight obtained through learning and the standard of said MLP if the reduction rate of a mean squared error converges within a permissible limit;
• and
• (c) reading the weight obtained in said step (b), receiving said inter-frame characteristic vectors from the currently analyzed frame, performing an operation of phoneme border discrimination to generate output values, discriminating the phoneme border according to the output values, and if the current analyzed frame arrives two frames preceding the final frame of the end portion of the vocal samples, outputting a frame number indicative of the border of phoneme as a final result.