• The Journal of the Acoustical Society of Korea
• /
• v.21 no.6
• /
• pp.522-530
• /
• 2002

This paper introduces a neural network (NN) -based nonlinear predictor for the LP (Linear Prediction) residual. To evaluate the effectiveness of the NN-based nonlinear predictor for LP-residual, we first compared the average prediction gain of the linear long-term predictor with that of the NN-based nonlinear long-term predictor. Then, the effects on the quantization noise of the nonlinear prediction residuals were investigated for the NN-based nonlinear predictor A new NN predictor takes into consideration not only prediction error but also quantization effects. To increase robustness against the quantization noise of the nonlinear prediction residual, a constrained back propagation learning algorithm, which satisfies a Kuhn-Tucker inequality condition is proposed. Experimental results indicate that the prediction gain of the proposed NN predictor was not seriously decreased even when the constrained optimization algorithm was employed.

• Proceedings of the KOSOMBE Conference
• /
• v.1992 no.05
• /
• pp.175-178
• /
• 1992

This paper proposes a method called linear prediction (a high performant technique in digital speech processing) for analyzing digital ECG signals. There are several significant properties indicating that ECG signals have an important feature in the residual error signal obtained after processing by Durbin's linear prediction algorithm. The ECG signal classification puts an emphasis on the residual error signal. For each ECG's QRS complex. the feature for recognition is obtained from a nonlinear transformation which transforms every residual error signal to set of three states pulse-cord train relative to the original ECG signal. The pulse-cord train has the advantage of easy implementation in digital hardware circuits to achive automated ECG diagnosis. The algorithm performs very well feature extraction in arrythmia detection. Using this method, our studies indicate that the PVC (premature ventricular contration) detection has a at least 90 percent sensityvity for arrythmia data.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.35S no.10
• /
• pp.166-174
• /
• 1998

Neural networks, having highly nonlinear dynamics by virtue of the distributed nonlinearities and the learing ability, have the potential for the adaptive prediction of nonstationary signals. This paper describes the nonlinear prediction of these signals in two ways; using a nonlinear module and the cascade combination of nonlinear and linear modules. Fully-connected recurrent neural networks (RNNs) and a conventional tapped-delay-line (TDL) filter are used as the nonlinear and linear modules respectively. The dynamic behavior of the proposed predictors is demonstrated for chaotic time series adn speech signals. For the relative comparison of prediction performance, the proposed predictors are compared with a conventional ARMA linear prediction model. Experimental results show that the neural networks based adaptive predictor ourperforms the traditional linear scheme significantly. We also find that the cascade combination predictor is well suitable for the prediction of the time series which contain large variations of signal amplitude.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.18 no.10
• /
• pp.1422-1432
• /
• 1993

Nonlinear mapping function of the HCNN( Hidden Control Neural Network ) can change over time to model the temporal variability of a speech signal by combining the nonlinear prediction of conventional neural networks with the segmentation capability of HMM. We have two things in this paper. first, we showed that the performance of the HCNN is better than that of HMM. Second, the HCNN with its prediction error measure given by weighted distance is proposed to use suitable distance measure for the HCNN, and then we showed that the superiority of the proposed system for speaker-independent speech recognition tasks. Weighted distance considers the differences between the variances of each component of the feature vector extraced from the speech data. Speaker-independent Korean digit recognition experiment showed that the recognition rate of 95%was obtained for the HCNN with Euclidean distance. This result is 1.28% higher than HMM, and shows that the HCNN which models the dynamical system is superior to HMM which is based on the statistical restrictions. And we obtained 97.35% for the HCNN with weighted distance, which is 2.35% better than the HCNN with Euclidean distance. The reason why the HCNN with weighted distance shows better performance is as follows : it reduces the variations of the recognition error rate over different speakers by increasing the recognition rate for the speakers who have many misclassified utterances. So we can conclude that the HCNN with weighted distance is more suit-able for speaker-independent speech recognition tasks.