한국어 고립단어 음성의 음성구간 검출 및 음가단위 자/모음 분류

이상범,이준환 단국대학교 1998 論文集 Vol.33 No.-

In this paper, Korean isolated word speech signal is analyzed suing Wavelet transform which has the similarity to human hearing model. After the multiresolution decomposition of speech signal using discrete Wavelet transform, the reconstruction signals of each frequency band are extracted by inverse Wavelet transform. For the reconstruction signals extracted, speech range detection is carried out using the feature parameter of energy. Also, after the extraction of energy, zero­crossing and level­crossing for th bandpass filtered reconstruction signals, phonetic value level Consonant/Vowel classification is carried out by organizing statistical pattern classifier for template patterns.

신경망을 이용한 서명 인식

李相範,姜成德,徐泳圭 단국대학교 1992 論文集 Vol.26 No.-

Recently, many algorithms for the image pattern recognition using the neural network have been proposed. In this paper, the characteristic parameters of signature pattern can be extracted by using discrete consine transform. The neural network is implemented with back propagation, generalized delta rule which is able to learn multi-layer perceptron among the various model and it is applied to the signature verification system as pattern classifier. Through the experimentation, the input of signature pattern which has N=16, is transformed with 2-dimension discrete cosine transform, and the characteristic parameters is applied to 3-layer perceptron neural network model. The experimental result shows the signature verification could be possible to verify them and the applied algorithms of neural network model could be applied to the signature verification.

Wavelet변환을 이용한 정지 영상 부호화 알고리즘

이상범,강성덕 단국대학교 1996 論文集 Vol.30 No.-

In order to encode image with efficiency, the time and spatial redundancy in the image should be removed without any information loss. This redundancy has been reduced by transforming image from the spatial domain to the frequency domain. In these transform schemes, lots of studies on Wavelet Transform have been done which is suitable to the human visual system and is able to reduce image redundancy efficiently. In this paper, among the second generation transform schemes which are thought to replace DCT transform, one of the most popular scheme in the image coding at the present, a still image coding scheme using Wavelet Transform has been proposed. In the still image coding scheme proposed in this paper, which enhanced the image compression method that employed Wavelet Transform and the VQ(vector quantization) suggested by Antonini, Wavelet Transform was repeated three times to separate the image by ten bands and it was tried to reduce the amount of data by eliminating the high frequency bands which rarely affect the human visual system out of the separated bands. And then still image compression using vector quantization was performed. In the vector quantization, MRC(multi-resolution codebook) appropriate for each band characteristics has been made up to reduce the image coding time, enhance the quality of the recovered image and increase the efficiency of the codebook by scanning the vector inputs vertically and horizontally to make them suitable to each band characteristics. As the result of this scanning, the PSNR of the recovered image has been improved by 0.5[db]. In the result of the study, the recovered image quality with thirty times as high compression rate and more than 30[db] has been achieved by assigning bits to each band efficiently.

LBG-VQ/HMM을 이용한 한국어 음성의 음가분할 및 인식에 관한 연구

이상범,이준환 단국대학교 2000 論文集 Vol.35 No.-

This paper investigates the Korean rule-based system and its speech segmentation application with phonetic value. The wavelet transform is used as a part of the front-end processor, by which phonetic value shaped extracts non-uniform blocks. In non-uniform blocks, extracted feature parameter is carried out by consonants/vowels phonetic value segmentation that is based on LBG-VQ/HMM. 4 speakers (male) evaluated the results on a portion of Korean isolated words typed TIMIT corpus consisting of the isolated Korean words.

FET 平衡 變調器의 設計

李相範 단국대학교 1980 論文集 Vol.14 No.-

This paper proposes a method for the design of an FET balanced modulator and illustrates the switching operation of the FET's from a set of modulation waveforms. The circuit inocrporates n-channel and p-channel FET such that no transformer is required. Direct coupling enables the modulator to operate over a wide frequency range and the circuit works equally well either as a suppressed carrier modulator or as an amplitude modulator with any desired amount of carrier. The various constraints of the signal has been prepared and the performance of that modulator has been checked by actual measurements.

근전도 신호 측정 장치의 개발

李相範,千在植,金昌福 단국대학교 1994 論文集 Vol.28 No.-

In this paper, electromyography measurement system is developed to measure the signal generated by the contraction and relaxation of the muscle from a living body. These signals gathered by the sensors are amplified by the amplifier, and then transferred to personal computer by multi-channel. Intel microprosser I8051 is used to control the data of electromyography. The analyzed electromyography signals can be displayed either in time domain and frequency domain. The data in frequency domain can be quickly converted into time domain by applying FFT(Fast Fourier Transform). Displays in time domain is adjustable according to the voltages and time scale.

한국어 음성 인식을 위한 컴퓨터 알고리즘의 개발

李相範,金聖珞,安昌 단국대학교 1989 論文集 Vol.23 No.-

Nowadays, speed recognition has based on speaker, vocabulary and so on. Many data, however, were required for speech analysis and synthesis. Thus it is necessary for the development of an algorithm to solve the problem. Since Korean syllables are divided into a consonant and vowel, a speech recognition algorithm using phonemes is effective. In this paper, 256 data samples and 12 predictor coefficients are used for the recognition. An effective voiced/unvoiced decision and pitch detection is performed by cepstrum. Also speech parameters are extracted by linear prediction and formant frequency analysis. Using these parameters, a speech recognition algorithm has been developed and make an experiment in some Korean speech samples. This experiment is performed for a speaker, and using linear prediction analysis and formant frequency extraction the acquired results are 87.14% and 88.09% respectively.

맥진 신호 측정 장치의 설계

이상범,강병욱 단국대학교 1998 論文集 Vol.32 No.-

In this paper, arterial pulse mesurement system is developed to measure the signal generated by the beating from a living body. These signals gathered by the sensors are amplified, and then are transferred to multi-channel of computer. Intel I8051 is used to control the data of the arterial pulse. The signal period of analized arterial pulse can be displayed either in time and frequency domain. The data in freqency domain can be quickly converted into time domain by applying FFT. Displays in time domain can be adjustable according to the voltage and scale.

펄스 信號의 에너지 帶域幅에 관한 考察

李相範 단국대학교 1981 論文集 Vol.15 No.-

Using a spectrum analyzer to measure the energy bandwidth (the bandwidth over which 90% of a pulse transmitter's energy is contained) doesn't always yield correct results, because this quantity is less than the main lobe displayed. The energy distribution is a function of two pulse characteristics that can be measured in real time with an oscilloscope, however; with this information and the aid of the curve shown here, energy bandwidth (EBW) can be found quickly and accurately.

單一칩 Processor를 利用한 BWDF Bank의 設計

李相範,林德奎 단국대학교 1986 論文集 Vol.20 No.-

An experimental implementation of a small digital filter bank on a commercially available single chip is studied. This processor is particularly suitable for implementation of wave digital structures since the shift·and-add arithmetic unit is very appropriate for exploiting the low coefficient word-length requirements of these structures. The measured results are in agreement with those expected and confirm the excellent stability properties of wave digital filter bank.