• International journal of advanced smart convergence
• /
• 제5권3호
• /
• pp.8-15
• /
• 2016

In recent years, one of deep learning models called Deep Belief Network (DBN) which formed by stacking restricted Boltzman machine in a greedy fashion has beed widely used for classification and recognition. With an ability to extracting features of high-level abstraction and deal with higher dimensional data structure, this model has ouperformed outstanding result on image and speech recognition. In this research, we assess the applicability of deep learning in dna classification level. Since the training phase of DBN is costly expensive, specially if deals with DNA sequence with thousand of variables, we introduce a new encoding method, using decimal-binary vector to represent the sequence as input to the model, thereafter compare with one-hot-vector encoding in two datasets. We evaluated our proposed model with different contrastive algorithms which achieved significant improvement for the training speed with comparable classification result. This result has shown a potential of using decimal-binary vector on DBN for DNA sequence to solve other sequence problem in bioinformatics.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2006년도 추계학술대회 발표논문집
• /
• pp.71-74
• /
• 2006

Many works have been done in the field of retrieving audio segments that contain human speeches without captions. To retrieve newly coined words and proper nouns, subwords were commonly used as indexing units in conjunction with query or document expansion. Among them, document expansion with subwords has serious drawback of large computation overhead. Therefore, in this paper, we propose Expected Matching Score based document expansion that effectively reduces computational overhead without much loss in retrieval precisions. Experiments have shown 13.9 times of speed up at the loss of 0.2% in the retrieval precision.

• 디지털산업정보학회논문지
• /
• 제10권1호
• /
• pp.131-138
• /
• 2014

This paper is on a stereo acoustic echo canceller with the blind signal separation for post processing. The convergence speed of the stereo acoustic echo canceller is deteriorated due to mixing two residual signals which are update signals of each echo canceller. To solve this problem, we are to use the blind signal separation(BSS) method separating the mixed signals after the echo cancellers. The blind signal separation method can extracts the source signals by means of the iterative computations with two input signals. We had verified performances of the proposed acoustic echo canceller for stereo through simulations. The results of simulations show that the acoustic echo canceller for stereo using this algorithm operates stably without divergence in the normal state. And, when the speech signals were inputted, this echo canceller achieved about 2dB higher ERLE with the BSS post processing method than without this method. This stereo echo canceller showed the best performance in the case of inputting the real voice signal.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1997년도 하계학술대회 논문집 A
• /
• pp.339-341
• /
• 1997

This study is presenting the method for an effective data transmission in MAGLEV which is now tested and intends to provide for an intelligent operation of signal system in future. To exchange a lot of information, it is ideal to adopt a digital system and a micro-based system is essential for these purposes. FSK modulation and HDLC protocol are adopted on this study and information line assembly which is used as the information exchange, as the speech communication, and as the detection of speed and position is constructed in one unit. Actually this study is produced academic achievements of the data transmission system of MAGLEV train and an advanced method of intelligent operation in future railway system.

• 전기학회논문지
• /
• 제60권2호
• /
• pp.423-430
• /
• 2011

In adaptive filtering, the sparseness of impulse response and input signal characteristics are very important factors of it's performance. This paper presents a subband improved proportionate normalized least square (SIPNLMS) algorithm which combines IPNLMS for impulse response sparseness and subband filtering for prewhitening the input signal. As drawing and combining the advantage of conventional approaches, the proposed algorithm, for impulse responses exhibiting high sparseness, achieve improved convergence speed and tracking ability. Simulation results, using colored signal(AR(4)) and speech input signals, show improved performance compared to fullband structure of existing methods.

• 디지털산업정보학회논문지
• /
• 제8권3호
• /
• pp.125-131
• /
• 2012

This paper is on a stereo acoustic echo canceller with the blind signal separation. The convergence speed of the stereo acoustic echo canceller is deteriorated due to mixing two residual signals in the update signal of each echo canceller. To solve this problem, we are to use the blind signal separation(BSS) method separating the mixed signals. The blind signal separation method can extracts the source signals by means of the iterative computations with two input signals. We had verified performances of the proposed acoustic echo canceller for stereo through simulations. The results of simulations show that the acoustic echo canceller for stereo using this algorithm operates stably without divergence in the normal state. And, when the speech signals were inputted, this echo canceller achieved about 3dB higher ERLE in the case of using the BSS algorithm than the case of not using the BSS algorithm. But this echo canceller didn't get good performances in the case of inputting the white noises as stereo signals.

• 대한음성학회지:말소리
• /
• 제58호
• /
• pp.83-99
• /
• 2006

One key factor that hinders the widespread deployment of speaker identification technologies is the requirement of long enrollment utterances to guarantee low error rate during identification. To gain user acceptance of speaker identification technologies, adaptation algorithms that can enroll speakers with short utterances are highly essential. To this end, this paper applies MLLR speaker adaptation for speaker enrollment and compares its performance against other speaker modeling techniques: GMMs and HMM. Also, to speed up the computational procedure of identification, we apply speaker clustering method which uses principal component analysis (PCA) and weighted Euclidean distance as distance measurement. Experimental results show that MLLR adapted modeling method is most effective for short enrollment utterances and that the GMMs performs better when long utterances are available.

• 한국정보과학회:학술대회논문집
• /
• 한국정보과학회 2004년도 봄 학술발표논문집 Vol.31 No.1 (B)
• /
• pp.571-573
• /
• 2004

Speaker verification systems based on multilayer perceptrons (MLPs) have good prospects in reliability and flexibility required as a successful authentication system. However, the poor learning speed of the error backpropagation (EBP) which is representative learning method of MLPs is the major defect to be complemented to achieve real-time user enrollments. In this paper, we implement an MLP-based speaker verification system and apply the existing two methods of the omitting patterns in instant learning (OIL) and the discriminative cohort speakers (DCS) to approach real-time enrollment. An evaluation of the system on a Korean speech database demonstrates the feasibility of the system as a speaker verification system of high performance.

• 전자통신동향분석
• /
• 제36권1호
• /
• pp.71-80
• /
• 2021

Artificial intelligence (AI) will have a huge impact on future education. We look at the role of AI in education and changes in schools. Personalized education is being attempted in limited services, and an interactive tutor service with speech recognition/dialog technology is being developed. In the future, we look forward to fully personalized education for individual students through AI teachers. Teachers are expected to make more effort to teach creative thinking, critical thinking, communication, and collaboration. As the speed of development of AI technology accelerates, we expect that AI-based education will be deeply established around us in the near future. We first introduce the details of the personalization technology and then discuss the AI-based foreign language speaking education research conducted by ETRI.

• 대한전자공학회논문지
• /
• 제23권4호
• /
• pp.429-438
• /
• 1986

In this paper we have designed and implemented a real-time, full-duplex SBC (sub-band coding) codec at 16kbps using a high speed digital signal processor, NEC 7720. The SBC codec employs a QMF(quadrature mirror filter) filter bank based on the tree structures of two-band analysis-synthesis pairs to partition speech signal into 4 octabe bands. Computer simulation has been done to investigate the effect of fixed-point computation of the NEC 7720. Three different performance measures, the conventional signal-to-noise ratio, the informal listening test, and an LPC(linear predictive coding)distance measure, have been used in this simulation. The necessary parameters have been optimized through the simulation. The developed hardware and software have been tested in real-time operation using a hardware emulator.