http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
장춘서 금오공과대학교 산업기술개발연구원 2001 産業技術開發硏究 Vol.17 No.-
In this paper, an effective speaker recognition method using recurrent TDNN(RTDNN) has been studied. RTDNN has been used to handle dynamic nature of each speaker. And this neural network with local feedback connection at the first hidden layer has been trained to extract representative feature vectors of the speaker. The suggested architecture shows improved speaker recognition results.
Microprogramming 기법에 의한 ADFE 실현 방식에 관한 연구
張椿瑞 金烏工科大學校 1987 論文集 Vol.8 No.-
In this paper, the effective architecture to implement an ADFE(adaptive decision feedback equalizer) is presented. This ADFE is designed by bit-slice microprocessor with 12bit wordlength and pipe line technique is used to increase micro-instruction execution speed. And for real time processing, a hardware multiplier is attached to the architecture. The overall ADFE system software is written by microprogram and to construct the system firmware efficiently the optimum micro-instruction definition has been considered. In addition, DPLL circuit for symbol timing recovery, symbol demodulation part and carrier tracking & adaptation part are presented.
32비트 부동소숫점 디지탈 신호처리 프로세서에 의한 멀티프로세서 구조에 관한 연구
張椿瑞,申允植 金烏工科大學校 1989 論文集 Vol.10 No.-
In this paper, the effective multi-processor architecture of 32 bit floating point digital signal processor(TMS 320C 30) has been studied. This multiprocessor system consists of one master module and several process modules. Master module can communicate with host computer or sample analog input signal and stores the samples in common memory. Each process module can access common memory module under the control of common bus arbitration logic and detailed circuits of this arbitration logic has been studied. In addition, local memory decode part, common memory decode part and wait signal generation logic have been suggestd.
장춘서 金烏工科大學校 1992 論文集 Vol.13 No.-
본 논문에서 한국어의 모든음소를 처리하여 단어레벨의 인식을 하기 위해 11개의 subclass신경망으로 구성된 모듈방식의 TDNN(time-delay neural network)의 구조에 대해 연구하였으며 이때 입력 프레임을 2 상태로 분할하여 음소 특성의 시간구조 처리를 가능하게하였다. 모듈 TDNN의 제2은닉층 2개는 FVQ(fuzzy vector quantizer)의 출력으로 사용되어 HMM의 입력이 되고 HMM의 알고리즘은 fuzzy HMM에 맞도록 변경되었다. 이와같은 혼성 신경회로망 구조를 통해 HMM과 TDNN의 장점이 결합되어 전체적인 음성인식 시스템의 성능향상을 기했으며 이를 컴퓨터 시뮬레이션을 통해 확인하였다.
장춘서 금오공과대학교 산업기술개발연구원 2001 産業技術開發硏究 Vol.17 No.-
In this paper, a new neural network architecture for speaker verification has been studied. To increase the ability to deal with the temporal structure of phonemic features, the input layer of the time-delay neural networks has been divided into several states. The optimum number of states has been proved by the result of computer simulation.
부동(浮動) 소숫점 디지탈 신호처리 프로세서에 의한 고속 실시간 FIR 필터링에 관한 연구
張椿瑞,李永淳 金烏工科大學校 1988 論文集 Vol.9 No.-
In this paper, the architecture of 32 bit floating-point digital signal processors(TMS 32030, DSP96000) and implementation of FIR filtering by these processors have been studied. Finite word length effect and overflow during the process of accumulation can be effectively eliminated by floating point digital signal processors. To increase the signal processing speed, these processors feature single cycle floating-point/integer multiplier, instruction cache or on-chip program RAM, multiple address buses and data buses, multiple address calculation units, extended precision registers and on-chip high speed memory block is suggested in this paper. In addition, circular buffering technique and multiply accamulate instruction with parallel data move have been suggested to simplify code implementation and to increase convolution speed.
장춘서,이기수,임은기 금오공과대학교 산업기술개발연구원 2000 産業技術開發硏究 Vol.16 No.-
In this paper, an effective phoneme recognition method for large vocabulary speech recognition system has been studied. VQ(Vector Quantization) is used as a front-end system for HMM(Hidden Markov Model) recognizer. Optimum number of phoneme classes and optimum size of code book for phoneme basis recognition environment has been proved by the result of computer simulation.
張椿瑞,李琦秀 金烏工科大學校 1988 論文集 Vol.9 No.-
In this paper, an effective hardware architecture to implement isolated word real-time speech recognizer has been studied. Filter bank analysis and VQ(Vector quantization) technique have been used to reduce speech processing time and DTW(dynamic time warping) has been used for time alignment. The hardware of speech recognizer consists of two parts : controller part ad DSP part. Controller part has a buffer control logic for common memory and SIO for interfacing to host computer. DSP part with 16bit 40 Mhz high speed processor(TMS 32025) performs recognition operation. In order to perform feature extraction and VQ encoding with in one frame, time requirement for data memory access and memory bank architecture have been suggested.
한국어 대용량 음성인식에서의 효율적인 시간 감축에 관한 연구
장춘서,이기수,신윤식,임은기,김병만,김현수 금오공과대학교 산업기술개발연구원 2000 産業技術開發硏究 Vol.16 No.-
In this paper a time reduction algorithm for large vocabulary speech recognition system has been studied. TDNN(Time Delay Neural Network) was used to classify vowel using phonetical fetures. To increase performance of vowel classification, modular TDNN architecture associated with 2 states input layer has been suggested.
頭鍼이 中風患者의 단기간 NIHSS와 MBI에 미치는 영향
류재춘,박인범,김상우,김철홍,서정철,윤현민,장경전,송춘호,안창범 동의대학교 한의학연구소 2002 동의한의연구 Vol.6 No.-
Objective ; The aim of this study was to investigate the effect of Scalp acupuncture on recovery of motor disorders in stroke patients. Methods ; Twenty two patients with poststroke-hemiplegia were randomized into two groups. Ten patients(test group) treated by 2 methods-Scalp acupuncture and body acupuncture. The other twelve patients(control group) treated only by body acupuncture. The activity of daily living was measured with a National Institutes of Health stroke scale(NIHSS) and Modified Barthel Index(MBI). The therapy was performed one a day for 2 weeks. Results ; In terms of score of NIHSS, the test group showed statistically meaningful decrease after 2 week treatment, but the control group showed statistically meaningful decrease after 1 week(p<0.05). And in terms of score of MBI, the test group did not show statistically meaningful increase but the control group showed statistically meaningful increase after 1 week(p<O.05). There was no statistically meaningful difference after 1 and 2 week treatment between the groups. Conclusions ; These results support that test group has almost same effectness compared with control group in improvement of the activity of daily living of poststroke-hemiplegic patients.