• MALSORI
• /
• no.59
• /
• pp.53-68
• /
• 2006

We report the evaluation results of the Korean speech recognition platform called ECHOS. The platform has an object-oriented and reusable architecture so that researchers can easily evaluate their own algorithms. The platform has all intrinsic modules to build a large vocabulary speech recognizer: Noise reduction, end-point detection, feature extraction, hidden Markov model (HMM)-based acoustic modeling, cross-word modeling, n-gram language modeling, n-best search, word graph generation, and Korean-specific language processing. The platform supports both lexical search trees and finite-state networks. It performs word-dependent n-best search with bigram in the forward search stage, and rescores the lattice with trigram in the backward stage. In an 8000-word continuous speech recognition task, the platform with a lexical tree increases 40% of word errors but decreases 50% of recognition time compared to the HTK platform with flat lexicon. ECHOS reduces 40% of recognition errors through incorporation of cross-word modeling. With the number of Gaussian mixtures increasing to 16, it yields word accuracy comparable to the previous lexical tree-based platform, Julius.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.8
• /
• pp.498-504
• /
• 2005

We introduce a Korean speech recognition platform (ECHOS) developed for education and research Purposes. ECHOS lowers the entry barrier to speech recognition research and can be used as a reference engine by providing elementary speech recognition modules. It has an easy simple object-oriented architecture, implemented in the C++ language with the standard template library. The input of the ECHOS is digital speech data sampled at 8 or 16 kHz. Its output is the 1-best recognition result. N-best recognition results, and a word graph. The recognition engine is composed of MFCC/PLP feature extraction, HMM-based acoustic modeling, n-gram language modeling, finite state network (FSN)- and lexical tree-based search algorithms. It can handle various tasks from isolated word recognition to large vocabulary continuous speech recognition. We compare the performance of ECHOS and hidden Markov model toolkit (HTK) for validation. In an FSN-based task. ECHOS shows similar word accuracy while the recognition time is doubled because of object-oriented implementation. For a 8000-word continuous speech recognition task, using the lexical tree search algorithm different from the algorithm used in HTK, it increases the word error rate by $40\%$ relatively but reduces the recognition time to half.

• Proceedings of the KIEE Conference
• /
• 1987.07b
• /
• pp.894-897
• /
• 1987

Listener echos, which arise in multiple 4-wire loop connections(MLC) during the evolving switched telephone network, impare voice-band data signal transmission performance. This paper first shows the calculation method of the total number of listener echo loops over N 4-wire physical loops and presents the additative law for listener echos. It next demonstrates that about 4 dB should be ensured to Eke the transmission loss of intraoffice call path be minimum for the voice-band data service in a digital local switch.

• Journal of Biomedical Engineering Research
• /
• v.14 no.2
• /
• pp.125-136
• /
• 1993

The study of Cochlear biomechanics is to clearly define three biomechanical principles of the Cochlea : Activity, Nonlinearity and Feedback. In this article, the Cochlea is linearly and actively modelled in one dimensional time domain. The sharp tunning of the Basilar Membrane displacement is shown when the amplifying activity of hair cells is added to the model. The amplified energy of the travelling displacement wave is emitted throughout the Cochlear fluid, so that the model becomes unstable. A new technique is introduced to reduce strong echos fro the Helicotrema. It makes the model less unstable. Both pure and click tones are used as input stimuli onto the ear durm. When the model is normal, the click response of the model shows that the backward emission of the amplified fluid pressure has mainly the echos from the Helicotrema. However, when the linear and active model is assumed to be abnormal, that is, some of hair cells are damaged not to produce the active process, the effect of the hair cell damage is resulted in the Oto-acoustic emission. The frequency response of the abnormally emitted sound pressure shows that the Oto-acoustic emission has the information about the characteristic frequency of the damaged hair cell. The main aim of this paper is to demonstrate the active biomechanics of the Chchlea in the time domain.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1985.10a
• /
• pp.47-49
• /
• 1985

In most of ultrasonic NDT(Non-Destructive Testing) equipments using A-scan display technic, it is one of the inconveniences that the user must be proficient in reading the displayed signals for the accurate decisions. In this study, a simple microprocessorized NDT machine for the flaw detection was developed. The operation of system is based on the conventional NDT system. The microprocessor detects the time delay between transmitted pulse and echos by counter-measure method. Then according to the scanning position, the location of flaw orthe other side of testing object is plotted on the CRT. The main advantages of the developed system are simplicity in handling, recording capability of measured data, and low cost.

• Journal of Ocean Engineering and Technology
• /
• v.13 no.4 s.35
• /
• pp.45-54
• /
• 1999

Non destructive test is applied to revise mechanical strength and assume material strength or defect of material, equipment and structure, instead of fracture test. Especially, ultrasonic test has the characteristics such as an excellent permeability high-sensitiveness to fine defect and an almost exact measurement for position, size and direction of inner defect which differ from other non destructive tests. In this study, the program is developed to evaluate optimal testing condition, to distinguish obstacle echo and defect position. This program on the basic of Ray-Tracing model shows generation and processing of ultrasonic pulse. The simulation is compared with testing in the 3 cases of an oblique angle transducer like $45^{\circ},\;60^{\circ}\;and\;70^{\circ}$. The test result for all conditions is well compared with simulation result when relative not is within $0.1{\sim}7.2%$. And the course of several echos is simply assumed through simulation.

• Proceedings of the Korean Institute of IIIuminating and Electrical Installation Engineers Conference
• /
• 2004.11a
• /
• pp.359-364
• /
• 2004

This paper deals with the comparison of signal processing techniques of ultrasonic data. The goal of signal processing is the ultrasonic speckle suppression and the visibility enhancement of flaw-reflected ultrasonic echo. The performance of conventional SSP(split spectrum processing) method and the wavelet denoising method are compared and discussed for tested ultrasonic data. Tested ultrasonic data obtained from the weld area of centrifugal-casted stainless steel material and safe-ending material with holes and notch of variable depths are presented. In experimental results, the outputs of wavelet-based denoising method show the clear and sharp peaks at the positions of flaw-reflected echos comparing with those of SSP method.

• Journal of the Korean Society of Manufacturing Technology Engineers
• /
• v.8 no.1
• /
• pp.135-141
• /
• 1999

Recently, advance signal analysis which is called "Time-Frequency Analysis" has been developed. Wavelet and Wigner Distribution are used to the method. Wavelet transform(WT) is applied to time-frequency analysis of waveforms obtained by an ultrasonic pulse-echo technique. The Gabor function is adopted as the analyzing wavelet. Wavelet analysis method is an attractive technique for evolution of material characterization evoluation. In this paper, the feasibility of suppression of ultrasonic background noise signal using WT has been presented. These results suggest that ultrasonic background noise ginal can be suppressed and enhanced even for SNR of 20.8 dB. This property of the WT is extremely useful for the detecting flaw echos embedded in background noise.und noise.

• Proceedings of the KIEE Conference
• /
• 1988.07a
• /
• pp.294-296
• /
• 1988

The tomographic imaging that employs ultrasonic echos has achieved outstanding advances in recent years, and today, ultrasonic diagnostic equipment has become the tool that is absolutely indispensible for clinical operations. Meanwhile, the feasility of measuring blood flow in the heart and vessels by the use of Doppler effect in ultrasonic waves is a well known fact. With respect to the method of blood flow measurment, there are two kinds which employ continous wave and pulse wave doppler system. In this paper, we describe the measurment of Blood flow-turbulence using general purpose Digital Signal Processing Board which had been implemented for the purpose of real-time spectrum analyser. Blood flow-turbulence means the blood-flow behavior. And it's value proportional to the spectrum variance. Therefore mean frequency of blood signal and variance provide useful diagnostic information. We have applied to the major arteries and vein, obtained the information about the time dependent blood-flow behavior.

• Proceedings of the KSPS conference
• /
• 2005.11a
• /
• pp.215-218
• /
• 2005

This paper reports the current status of development of the Korean speech recognition platform (ECHOS). We implement new modules including ETSI feature extraction, backward search with trigram, and utterance verification. The ETSI feature extraction module is implemented by converting the public software to an object-oriented program. We show that trigram language modeling in the backward search pass reduces the word error rate from 23.5% to 22% on a large vocabulary continuous speech recognition task. We confirm the utterance verification module by examining word graphs with confidence score.