• Journal of the Korean Data and Information Science Society
• /
• v.28 no.6
• /
• pp.1217-1227
• /
• 2017

Deep learning has been studied explosively and has achieved excellent performance in areas like image and speech recognition, the application areas in which computations have been challenges with ordinary machine learning techniques. The theoretical study of deep learning has also been researched toward improving the performance. In this paper, we try to find a key of the success of the deep learning in rich and efficient expressiveness of the deep learning function, and analyze the theoretical studies related to it.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.28B no.5
• /
• pp.394-403
• /
• 1991

This paper described the development of input driven multilayer perceptron(IDMLP) neural network and it's application to the Korean spoken digit recognition. The IDMPLP neural network used here and the learning algorithm for this network was proposed newly. In this model, weight value is integer and transfer function in the neuron is hard limit function. According to the result of the network learning for the some kinds of input data, the number of network layers is one or more by the difficulties of classifying the inputs. We tested the recognition of binaried data for the spoken digit 0 to 9 by means of the proposed network. The experimental results are 100% and 96% for the learning data and test data, respectively.

• Journal of the Korean Institute of Telematics and Electronics C
• /
• v.34C no.1
• /
• pp.42-50
• /
• 1997

Vector quantization has shown to be useful for compressing data related with a wide rnage of applications such as image processing, speech processing, and weather satellite. Neural networks of images this paper propses a efficient neural network learning algorithm, called learning count control algorithm based on the frquency sensitive learning algorithm. This algorithm can train a results more codewords can be assigned to the sensitive region of the human visual system and the quality of the reconstructed imate can be improved. We use a human visual systrem model that is a cascade of a nonlinear intensity mapping function and a modulation transfer function with a bandpass characteristic.

• Proceedings of the Korea Society for Industrial Systems Conference
• /
• 2003.11a
• /
• pp.721-727
• /
• 2003

This study designed and implemented a system to extract of vowel from a word. The system is comprised of a voice feature extraction module and a neutral network module. The voice feature extraction module use a LPC(Linear Prediction Coefficient) model to extract a voice feature from a word. The neutral network module is comprised of a learning module and voice recognition module. The learning module sets up a learning pattern and builds up a neutral network to learn. Using the information of a learned neutral network, a voice recognition module extracts a vowel from a word. A neutral network was made to learn selected vowels(a, eo, o, e, i) to test the performance of a implemented vowel extraction recognition machine. Through this experiment, could confirm that speech recognition module extract of vowel from 4 words.

• Journal of Biomedical Engineering Research
• /
• v.45 no.1
• /
• pp.1-9
• /
• 2024

Communication with others plays an important role in human social interaction and information exchange in modern society. However, some individuals have difficulty in communicating due to dysarthria. Therefore, it is necessary to develop effective diagnostic techniques for early treatment of the dysarthria. In the present study, we propose a mobile device-based methodology that enables to automatically classify dysarthria type. The light-weight CNN model was trained by using the open audio dataset of Korean patients with dysarthria. The trained CNN model can successfully classify dysarthria into related subtype disease with 78.8%~96.6% accuracy. In addition, the user-friendly mobile application was also developed based on the trained CNN model. Users can easily record their voices according to the selected inspection type (e.g. word, sentence, paragraph, and semi-free speech) and evaluate the recorded voice data through their mobile device and the developed mobile application. This proposed technique would be helpful for personal management of dysarthria and decision making in clinic.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.3
• /
• pp.469-475
• /
• 2022

To train the model of the deep learning-based single-speaker TTS system, a speech DB of tens of hours and a lot of training time are required. This is an inefficient method in terms of time and cost to train multi-speaker or personalized TTS models. The voice cloning method uses a speaker encoder model to make the TTS model of a new speaker. Through the trained speaker encoder model, a speaker embedding vector representing the timbre of the new speaker is created from the small speech data of the new speaker that is not used for training. In this paper, we propose a multi-speaker TTS system to which voice cloning is applied. The proposed TTS system consists of a speaker encoder, synthesizer and vocoder. The speaker encoder applies the d-vector technique used in the speaker recognition field. The timbre of the new speaker is expressed by adding the d-vector derived from the trained speaker encoder as an input to the synthesizer. It can be seen that the performance of the proposed TTS system is excellent from the experimental results derived by the MOS and timbre similarity listening tests.

• Journal of Digital Convergence
• /
• v.17 no.7
• /
• pp.285-292
• /
• 2019

With the popularization of PC, SNS and IoT, a lot of data is generated and the amount is increasing exponentially. Artificial neural network learning is a topic that attracts attention in many fields in recent years by using huge amounts of data. Artificial neural network learning has shown tremendous potential in speech recognition and image recognition, and is widely applied to a variety of complex areas such as medical diagnosis, artificial intelligence games, and face recognition. The results of artificial neural networks are accurate enough to surpass real human beings. Despite these many advantages, privacy problems still exist in artificial neural network learning. Learning data for artificial neural network learning includes various information including personal sensitive information, so that privacy can be exposed due to malicious attackers. There is a privacy risk that occurs when an attacker interferes with learning and degrades learning or attacks a model that has completed learning. In this paper, we analyze the attack method of the recently proposed neural network model and its privacy protection method.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.2
• /
• pp.176-182
• /
• 2021

This paper evaluates and compares the performance of the Deep Nerual Network (DNN)-based speech enhancement models according to various loss functions. We used a complex network that can consider the phase information of speech as a baseline model. As the loss function, we consider two types of basic loss functions; the Mean Squared Error (MSE) and the Scale-Invariant Source-to-Noise Ratio (SI-SNR), and two types of perceptual-based loss functions, including the Perceptual Metric for Speech Quality Evaluation (PMSQE) and the Log Mel Spectra (LMS). The performance comparison was performed through objective evaluation and listening tests with outputs obtained using various combinations of the loss functions. Test results show that when a perceptual-based loss function was combined with MSE or SI-SNR, the overall performance is improved, and the perceptual-based loss functions, even exhibiting lower objective scores showed better performance in the listening test.

• Speech Sciences
• /
• v.9 no.1
• /
• pp.89-110
• /
• 2002

Intonation is very difficult to describe and it is furthermore difficult to compare intonation between different languages because of their differences of intonation systems. This paper aims to compare some intonation phenomena between Korean, French and English. In this paper I will refer to ToBI (the Tone and Break Indices) which is a prosodic transcription model proposed originally by Pierrehumbert (1980) as a description tool. In the first part, I will summarize different ToBI systems, namely, K-ToBI (Korean ToBI), F-ToBI (French ToBI) and ToBI itself (English ToBI) in order to compare the differences of three languages within prosody. In the second part, I will analyze some tokens registered by Korean, French and American in different languages to show the difficulties of learning other languages and to find the prosodic cues to pronounce correctly other languages. The point of comparison in this study is the Accentual Phrase (AP) in Korean and in French and the intermediate phrase (ip) in English, which I will call ' subject phrase ' in this study for convenience.

• Journal of Information Science Theory and Practice
• /
• v.2 no.3
• /
• pp.29-39
• /
• 2014

In recent years, several studies have proposed making use of the Twitter micro-blogging service to track various trends in online media and discussion. In this study, we specifically examine the use of Twitter to track discussions of food safety in the Korean language. Given the irregularity of keyword use in most tweets, we focus on optimistic machine-learning and feature set selection to classify collected tweets. We build the classifier model using Naive Bayes & Naive Bayes Multinomial, Support Vector Machine, and Decision Tree Algorithms, all of which show good performance. To select an optimum feature set, we construct a basic feature set as a standard for performance comparison, so that further test feature sets can be evaluated. Experiments show that precision and F-measure performance are best when using a Naive Bayes Multinomial classifier model with a test feature set defined by extracting Substantive, Predicate, Modifier, and Interjection parts of speech.