• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.34 no.3
• /
• pp.439-449
• /
• 2024

Recent rapid advancements in voice generation technology have enabled the natural synthesis of voices using text alone. However, this progress has led to an increase in malicious activities, such as voice phishing (voishing), where generated voices are exploited for criminal purposes. Numerous models have been developed to detect the presence of synthesized voices, typically by extracting features from the voice and using these features to determine the likelihood of voice generation.This paper proposes a new model for extracting voice features to address misuse cases arising from generated voices. It utilizes a deep learning-based audio codec model and the pre-trained natural language processing model BERT to extract novel voice features. To assess the suitability of the proposed voice feature extraction model for voice detection, four generated voice detection models were created using the extracted features, and performance evaluations were conducted. For performance comparison, three voice detection models based on Deepfeature proposed in previous studies were evaluated against other models in terms of accuracy and EER. The model proposed in this paper achieved an accuracy of 88.08%and a low EER of 11.79%, outperforming the existing models. These results confirm that the voice feature extraction method introduced in this paper can be an effective tool for distinguishing between generated and real voices.

• Journal of Korea Multimedia Society
• /
• v.7 no.10
• /
• pp.1452-1463
• /
• 2004

The requirement of automatic video summary is increasing as bi-directional broadcasting contents and various user requests and preferences for the bi -directional broadcast environment are increasing. Automatic video summary is needed for an efficient management and usage of many contents in service provider as well. In this paper, we propose a method to generate a content-based summary of linguistic educational videos automatically. First, shot-boundaries and keyframes are generated from linguistic educational video and then multiple(low-level) visual features are extracted. Next, the semantic parts (Explanation part, Dialog part, Text-based part) of the linguistic educational video are generated using extracted visual features. Lastly the XMI- document describing summary information is made based on HieraTchical Summary architecture oi MPEG-7 MDS (Multimedia I)escription Scheme). Experimental results show that our proposed algorithm provides reasonable performance for automatic summary of linguistic educational videos. We verified that the proposed method is useful ior video summary system to provide various services as well as management of educational contents.

• The KIPS Transactions:PartD
• /
• v.9D no.5
• /
• pp.779-788
• /
• 2002

There are many technically difficult issues in searching multimedia data such as image, video and audio because they are massive and more complex than simple text-based data. As a method of searching multimedia data, a similarity retrieval has been studied to retrieve automatically basic features of multimedia data and to make a search among data with retrieved features because exact match is not adaptable to a matrix of features of multimedia. In this paper, data clustering and its indexing are proposed as a speedy similarity-retrieval method of multimedia data. This approach clusters similar images on adjacent disk cylinders and then builds Indexes to access the clusters. To minimize the search cost, the hashing is adapted to index cluster. In addition, to reduce I/O time, the proposed searching takes just one I/O to look up the location of the cluster containing similar object and one sequential file I/O to read in this cluster. The proposed schema solves the problem of multi-dimension by using clustering and its indexing and has higher search efficiency than the content-based image retrieval that uses only clustering or indexing structure.

• Journal of Fashion Business
• /
• v.11 no.5
• /
• pp.110-123
• /
• 2007

This study aims at analyzing the characteristics and the mode of deconstructionism expressed in fashion design centering on review of literatures and precedent researches. Through the review of previous studies and various articles, three major features with regard to deconstructionism were found: indeterminacy of meaning, decentralization, and inter -textuality. In fashion, the feature of indeterminacy of meaning expressed by non formula way such as unfinished, layering, obi or straight line cutting and it was divided into unstructured and unorganized factors. The feature of decentralization expressed by disorganization of sex, race, culture and humanism such as trans-gender, third world costume, sub-culture, posthumanism. Last, the feature of inter-textuality expressed by employing the fashion texts more than two simultaneously such as mixture of unmatched or unusual various clothing items, fabrics, patterns and styles. Non formality features in fahsion which divided into unstructured and unconstructed factors. The decentering of meaning means that disorganized the existing concepts of sex, race, culture and humanism features in fashion which divided into cross-gender, subculture, and post-humanism. Lastly, the feature of inter-textuality means that employed the fashion text more than two which is classified into factors of mixing mode and mixed styles. Unstructured factors were expressed by destroying the design principles such as balance, harmony and unit in fashion. Unconstructed factors were expressed by ignoring the basic costume structures and components in fashion. Cross-gender was expressed by manish and duality style in fashion and 3rd world costume was expressed by folk costume style. Anti fashion was expressed by anti aesthetics in fashion such as kitch, punk, and hippie styles. Post-human was expressed by futurism and the 4th Sex style in fashion. Mixing mode were expressed by mixture of item, and unmatched materials and pattern in fashion. Mixture of style were expressed by mixture of pastiche expression of various styles.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.35 no.11C
• /
• pp.929-939
• /
• 2010

The growth of multimedia technology forces the development of video detection for large database management and illegal copy detection. To meet this demand, this paper proposes a fast video detection method to apply to a large database. The fast video detection algorithm uses spatial features using the gray value distribution from frames and temporal features using the temporal similarity map. We form the video signature using the extracted spatial feature and temporal feature, and carry out a stepwise matching method. The performance was evaluated by accuracy, extraction and matching time, and signature size using the original videos and their modified versions such as brightness change, lossy compression, text/logo overlay. We show empirical parameter selection and the experimental results for the simple matching method using only spatial feature and compare the results with existing algorithms. According to the experimental results, the proposed method has good performance in accuracy, processing time, and signature size. Therefore, the proposed fast detection algorithm is suitable for video detection with the large database.

• Journal of the Korea Safety Management & Science
• /
• v.14 no.4
• /
• pp.229-238
• /
• 2012

Within last 10 years, internet has become a daily activity, and humankind had to face the Data Deluge, a dramatic increase of digital data (Economist 2012). Due to exponential increase in amount of digital data, large scale data has become a big issue and hence the term 'big data' appeared. There is no official agreement in quantitative and detailed definition of the 'big data', but the meaning is expanding to its value and efficacy. Big data not only has the standardized personal information (internal) like customer information, but also has complex data of external, atypical, social, and real time data. Big data's technology has the concept that covers wide range technology, including 'data achievement, save/manage, analysis, and application'. To define the connected technology of 'big data', there are Big Table, Cassandra, Hadoop, MapReduce, Hbase, and NoSQL, and for the sub-techniques, Text Mining, Opinion Mining, Social Network Analysis, Cluster Analysis are gaining attention. The three features that 'bid data' needs to have is about creating large amounts of individual elements (high-resolution) to variety of high-frequency data. Big data has three defining features of volume, variety, and velocity, which is called the '3V'. There is increase in complexity as the 4th feature, and as all 4features are satisfied, it becomes more suitable to a 'big data'. In this study, we have looked at various reasons why companies need to impose 'big data', ways of application, and advanced cases of domestic and foreign applications. To correspond effectively to 'big data' revolution, paradigm shift in areas of data production, distribution, and consumption is needed, and insight of unfolding and preparing future business by considering the unpredictable market of technology, industry environment, and flow of social demand is desperately needed.

• Journal of Multimedia Information System
• /
• v.9 no.2
• /
• pp.113-120
• /
• 2022

Mongolian folk songs are inspired by Mongolian labor songs and are classified into long and short songs. Mongolian long songs have ancient origins, are rich in legends, and are a great source of folklore. So it was inscribed by UNESCO in 2008. Mongolian written literature is formed under the direct influence of oral literature. Mongolian long song has 3 classes: ayzam, suman, and besreg by their lyrics and structure. In ayzam long song, the world perfectly embodies the philosophical nature of world phenomena and the nature of human life. Suman long song has a wide range of topics such as the common way of life, respect for ancestors, respect for fathers, respect for mountains and water, livestock and animal husbandry, as well as the history of Mongolia. Besreg long songs are dominated by commanded and trained characters. In this paper, we proposed a method to classify their 3 types of long songs using machine learning, based on their lyrics structures without semantic information. We collected lyrics of over 80 long songs and extracted 11 features from every single song. The features are the name of a song, number of the verse, number of lines, number of words, general value, double value, elapsed time of verse, elapsed time of 5 words, and the longest elapsed time of 1 word, full text, and type label. In experimental results, our proposed features show on average 78% recognition rates in function type machine learning methods, to classify the ayzam, suman, and besreg classes.

• Journal of Intelligence and Information Systems
• /
• v.25 no.4
• /
• pp.105-122
• /
• 2019

Dimensionality reduction is one of the methods to handle big data in text mining. For dimensionality reduction, we should consider the density of data, which has a significant influence on the performance of sentence classification. It requires lots of computations for data of higher dimensions. Eventually, it can cause lots of computational cost and overfitting in the model. Thus, the dimension reduction process is necessary to improve the performance of the model. Diverse methods have been proposed from only lessening the noise of data like misspelling or informal text to including semantic and syntactic information. On top of it, the expression and selection of the text features have impacts on the performance of the classifier for sentence classification, which is one of the fields of Natural Language Processing. The common goal of dimension reduction is to find latent space that is representative of raw data from observation space. Existing methods utilize various algorithms for dimensionality reduction, such as feature extraction and feature selection. In addition to these algorithms, word embeddings, learning low-dimensional vector space representations of words, that can capture semantic and syntactic information from data are also utilized. For improving performance, recent studies have suggested methods that the word dictionary is modified according to the positive and negative score of pre-defined words. The basic idea of this study is that similar words have similar vector representations. Once the feature selection algorithm selects the words that are not important, we thought the words that are similar to the selected words also have no impacts on sentence classification. This study proposes two ways to achieve more accurate classification that conduct selective word elimination under specific regulations and construct word embedding based on Word2Vec embedding. To select words having low importance from the text, we use information gain algorithm to measure the importance and cosine similarity to search for similar words. First, we eliminate words that have comparatively low information gain values from the raw text and form word embedding. Second, we select words additionally that are similar to the words that have a low level of information gain values and make word embedding. In the end, these filtered text and word embedding apply to the deep learning models; Convolutional Neural Network and Attention-Based Bidirectional LSTM. This study uses customer reviews on Kindle in Amazon.com, IMDB, and Yelp as datasets, and classify each data using the deep learning models. The reviews got more than five helpful votes, and the ratio of helpful votes was over 70% classified as helpful reviews. Also, Yelp only shows the number of helpful votes. We extracted 100,000 reviews which got more than five helpful votes using a random sampling method among 750,000 reviews. The minimal preprocessing was executed to each dataset, such as removing numbers and special characters from text data. To evaluate the proposed methods, we compared the performances of Word2Vec and GloVe word embeddings, which used all the words. We showed that one of the proposed methods is better than the embeddings with all the words. By removing unimportant words, we can get better performance. However, if we removed too many words, it showed that the performance was lowered. For future research, it is required to consider diverse ways of preprocessing and the in-depth analysis for the co-occurrence of words to measure similarity values among words. Also, we only applied the proposed method with Word2Vec. Other embedding methods such as GloVe, fastText, ELMo can be applied with the proposed methods, and it is possible to identify the possible combinations between word embedding methods and elimination methods.

• Journal of Intelligence and Information Systems
• /
• v.24 no.3
• /
• pp.45-66
• /
• 2018

Crowdfunding has become more popular than angel funding for fundraising by venture companies. Identification of success factors may be useful for fundraisers and investors to make decisions related to crowdfunding projects and predict a priori whether they will be successful or not. Recent studies have suggested several numeric factors, such as project goals and the number of associated SNS, studying how these affect the success of crowdfunding campaigns. However, prediction of the success of crowdfunding campaigns via non-numeric and unstructured data is not yet possible, especially through analysis of structural characteristics of documents introducing projects in need of funding. Analysis of these documents is promising because they are open and inexpensive to obtain. We propose a novel method to predict the success of a crowdfunding project based on the introductory text. To test the performance of the proposed method, in our study, texts related to 1,980 actual crowdfunding projects were collected and empirically analyzed. From the text data set, the following details about the projects were collected: category, number of replies, funding goal, fundraising method, reward, number of SNS followers, number of images and videos, and miscellaneous numeric data. These factors were identified as significant input features to be used in classification algorithms. The results suggest that the proposed method outperforms other recently proposed, non-text-based methods in terms of accuracy, F-score, and elapsed time.

• Science of Emotion and Sensibility
• /
• v.17 no.1
• /
• pp.71-78
• /
• 2014

We propose the sentiment pattern as a novel sentiment feature for more accurate text sentiment analysis, and introduce the rating inference of movie reviews using it. The text sentiment analysis is a task that recognizes and classifies sentiment of text whether it is positive or negative. For that purpose, the sentiment feature is used, which includes sentiment words and phrase pattern that have specific sentiment like positive or negative. The previous researches for the sentiment analysis, however, have a limit to understand accurately total sentiment of either a sentence or text because they consider the sentiment of sentiment words and phrase patterns independently. Therefore, we propose the sentiment pattern that is defined by arranging semantically all sentiment in a sentence, and use them as a new sentiment feature for the rating inference that is one of the detail subjects of the sentiment analysis. In order to verify the effect of proposed sentiment pattern, we conducted experiments of rating inference. Ratings of test reviews is inferred by using a probabilistic method with sentiment features including sentiment patterns extracted from training reviews. As a result, it is shown that the result of rating inference with sentiment patterns are more accurate than that without sentiment patterns.