• The Journal of the Acoustical Society of Korea
• /
• v.34 no.4
• /
• pp.316-320
• /
• 2015

In this paper, the pole filtering concept is applied to the Mel-frequency cepstral coefficient (MFCC) feature vectors in the conventional cepstral mean normalization (CMN) and cepstral mean and variance normalization (CMVN) frameworks. Additionally, performance of the cepstral mean and scale normalization (CMSN), which uses scale normalization instead of variance normalization, is evaluated in speech recognition experiments in noisy environments. Because CMN and CMVN are usually performed on a per-utterance basis, in case of short utterance, they have a problem that reliable estimation of the mean and variance is not guaranteed. However, by applying the pole filtering and scale normalization techniques to the feature normalization process, this problem can be relieved. Experimental results using Aurora 2 database (DB) show that feature normalization method combining the pole-filtering and scale normalization yields the best improvements.

• Journal of the Korea Society of Computer and Information
• /
• v.20 no.1
• /
• pp.21-28
• /
• 2015

Feature normalization as a pre-processing step has been widely used to reduce the effect of different scale in each feature dimension and error rate in classification. Most of the existing normalization methods, however, do not use the class labels of data points and, as a result, do not guarantee the optimality of normalization in classification aspect. A supervised rank normalization method, combination of rank normalization and supervised learning technique, was proposed and demonstrated better result than others. In this paper, another technique, training sample selection, is introduced in supervised feature normalization to reduce classification error more. Training sample selection is a common technique for increasing classification accuracy by removing noisy samples and can be applied in supervised normalization method. Two sample selection measures based on the classes of neighboring samples and the distance to neighboring samples were proposed and both of them showed better results than previous supervised rank normalization method.

• Communications for Statistical Applications and Methods
• /
• v.31 no.1
• /
• pp.105-127
• /
• 2024

We propose recurrent neural network (RNN) methods for forecasting realized volatility (RV). The data are RVs of ten major stock price indices, four from the US, and six from the EU. Forecasts are made for relative ratio of adjacent RVs instead of the RV itself in order to avoid the out-of-scale issue. Forecasts of RV ratios distribution are first constructed from which those of RVs are computed which are shown to be better than forecasts constructed directly from RV. The apparent asymmetry of RV ratio is addressed by the Piecewise Min-max (PM) normalization. The serial dependence of the ratio data renders us to consider two architectures, long short-term memory (LSTM) and gated recurrent unit (GRU). The hyperparameters of LSTM and GRU are tuned by the nested cross validation. The RNN forecast with the PM normalization and ratio transformation is shown to outperform other forecasts by other RNN models and by benchmarking models of the AR model, the support vector machine (SVM), the deep neural network (DNN), and the convolutional neural network (CNN).

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.11
• /
• pp.31-38
• /
• 2013

Feature normalization as a pre-processing step has been widely used in classification problems to reduce the effect of different scale in each feature dimension and error as a result. Most of the existing methods, however, assume some distribution function on feature distribution. Even worse, existing methods do not use the labels of data points and, as a result, do not guarantee the optimality of the normalization results in classification. In this paper, proposed is a supervised rank normalization which combines rank normalization and a supervised learning technique. The proposed method does not assume any feature distribution like rank normalization and uses class labels of nearest neighbors in classification to reduce error. SVM, in particular, tries to draw a decision boundary in the middle of class overlapping zone, the reduction of data density in that area helps SVM to find a decision boundary reducing generalized error. All the things mentioned above can be verified through experimental results.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.14 no.3
• /
• pp.547-552
• /
• 2019

Recently, the massive growth in the scale of data has been observed as a major issue in the Big Data. Furthermore, the Big Data should be preprocessed for normalization to get a high performance of the Machine learning since the Big Data is also an input of Machine Learning. The performance varies by many factors such as the scope of the columns in a Big Data or the methods of normalization preprocessing. In this paper, the various types of normalization preprocessing methods and the scopes of the Big Data columns will be applied to the SVM(: Support Vector Machine) as a Machine Learning method to get the efficient environment for the normalization preprocessing. The Machine Learning experiment has been programmed in Python and the Jupyter Notebook.

• Journal of Korean Society of Transportation
• /
• v.14 no.1
• /
• pp.135-154
• /
• 1996

This paper presents a heuristic algorithm for classifying vehicles using a single loop detector. The data used for the development of the algorithm are the frequency variation of a vehicle sensored from the circle-shaped loop detectors which are normal buried beneath the expressway. The pre-processing of data is required for the development of the algorithm that actually consists of two parts. One is both normalization of occupancy time and that with frequency variation, the other is finding of an adaptable number of sample size for each vehicle category and calculation of average value of normalized frequencies along with occupancy time that will be stored for comparison. Then, detected values are compared with those stored data to locate the most fitted pattern. After the normalization process, we developed some frameworks for comparison schemes. The fitted scales used were 10 and 15 frames in occupancy time(X-axis) and 10 and 15 frames in frequency variation (Y-axis). A combination of X-Y 10-15 frame turned out to be the most efficient scale of normalization producing 96 percent correct classification rate for six types of vehicle.

• Journal of Electrical Engineering and information Science
• /
• v.2 no.6
• /
• pp.146-155
• /
• 1997

In order to achieve scale- and rotation-invariance in recognizing unoccluded objects in binary images using Zernike moment features, an image of an object has often been normalized first by its zeroth-order moment (ZOM) or area. With elongated objects such as characters, a stroke width varies with the threshold value used, it becomes one or two pixels wider or thinner. The variations of the total area of the character becomes significant when the character is relatively thin with respect to its overall size, and the resulting normalized moment features are no longer reliable. This dilation/erosion effect is more severe when the object is not focused precisely. In this paper, we analyze the ZOM method and propose as a normalization method, the maximum enclosing circle (MEC) centered at the centroid of the character. We compare both the ZOM and MEC methods in their performance through various experiments.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.24 no.5
• /
• pp.905-913
• /
• 1987

This paper presents a speaker dependent isolated digit recognition algorithm using the Mellin transform. Since the Mellin transform converts a scale information into a phase information, attempts have been made to utilize this scale invariance property of the Mellin transform in order to alleviate a time-normalization procedure required for a speech recognition. It has been found that good results can be obtained by taking the Mellin transform to the features such as a ZCR, log energy, normalized autocorrelation coefficients, first predictor coefficient and normalized prediction error. We employed a difference function for evaluating a similarity between two patterns. When the proposed algorithm was tested on Korean digit words, a recognition rate of 83.3% was obtained. The recognition accuracy is not compatible with the other technique such as LPC distance however, it is believed that the Mellin transform can effectively perform the time-normalization processing for the speech recognition.

• Proceedings of the Korean Statistical Society Conference
• /
• 2002.05a
• /
• pp.175-181
• /
• 2002

cDNA microarray experiments permit us to investigate the expression levels of thousands of genes simultaneously and to make it easy to compare gene expression from different populations. However, researchers are asked to be cautious in interpreting the results because of the unexpected sources of variation such as systematic errors from the microarrayer and the difference of cDNA dye intensity. And the scanner itself calculates both of mean and median of the signal and background pixels, so it follows a selection which raw data will be used in analysis. In this paper, we compare the results in each case of using mean and median from the raw data and normalization methods in reducing the systematic errors with arm's skin cells of old and young males. Using median is preferable to mean because the distribution of the test statistic (t-statistic) from the median is more close to normal distribution than that from mean. Scaled print tip normalization is better than global or lowess normalization due to the distribution of the test-statistic.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.7 no.7
• /
• pp.1380-1386
• /
• 2003

Speech recognition performance is degraded by the variation in vocal tract length among speakers. In this paper, we have used a vocal tract length normalization method wherein the frequency axis of the short-time spectrum associated with a speaker's speech is scaled to minimize the effects of speaker's vocal tract length on the speech recognition performance In order to normalize vocal tract length, we tried several frequency warping functions such as linear and piece-wise linear function. Variable interval piece-wise linear warping function is proposed to effectively model the variation of frequency axis scale due to the large variation of vocal tract length. Experimental results on TIDIGITS connected digits showed the dramatic reduction of word error rates from 2.15% to 0.53% by the proposed vocal tract normalization.