• Proceedings of the IEEK Conference
• /
• 1999.11a
• /
• pp.368-371
• /
• 1999

In this paper, we propose a neural network processor architecture with on-chip learning and with reconfigurability according to the data dependencies of the algorithm applied. For the neural network model applied, the proposed architecture can be configured into either SIMD or SRA(Systolic Ring Array) without my changing of on-chip configuration so as to obtain a high throughput. However, changing of system configuration can be controlled by user program. To process activation function, which needs amount of cycles to get its value, we design it by using PWL(Piece-Wise Linear) function approximation method. This unit has only single latency and the processing ability of non-linear function such as sigmoid gaussian function etc. And we verified the processing mechanism with EBP(Error Back-Propagation) model.

• Phonetics and Speech Sciences
• /
• v.6 no.4
• /
• pp.127-132
• /
• 2014

This paper describes a robust speech recognition technique by reconstructing spectral components mismatched with a training environment. Although the cluster-based reconstruction method can compensate the unreliable components from reliable components in the same spectral vector by assuming an independent, identically distributed Gaussian-mixture process of training spectral vectors, the presented method exploits the temporal dependency of speech to reconstruct the components by introducing a hidden-Markov-model prior which incorporates an internal state transition plausible for an observed spectral vector sequence. The experimental results indicate that the described method can provide temporally consistent reconstruction and further improve recognition performance on average compared to the conventional method.

• Journal of Biomedical Engineering Research
• /
• v.15 no.3
• /
• pp.305-316
• /
• 1994

All physical data in the real world are nonstationary signals that have the time varying statistical characteristics. Although few algorithms suitable to process the nonstationary signals have ever been suggested, these are treated the nonstationary signals under the assumption that the nonstationary signal is a piece-wise stationary signal. Recently, statistical analysis algorithms for the nonstationary signal have concentrated so much interest. In this paper, nonstationary EMG signals are mapped onto the orthogonal wavelet transform domain so that the eigenvalue spread of its autocorrelation matrix could be more smaller than that in the time domain. Then the model in the wavelet transform domain and an algorithm to estimate the model parameters are suggested. Also, an test signal generated by a white gaussian noise and the EMG signal are identified, and the algorithm performance is considered in the sense of the mean square error and the evaluation parameters.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2003.06a
• /
• pp.18-21
• /
• 2003

In general it has been considered to be the difficult problem that we divide continuous speech into short interval with having identical phoneme quality. In this paper we used Gaussian Mixture Model (GMM) related to probability density to divide speech into phonemes, an initial, medial, and final sound. From them we peformed continuous speech recognition. Decision boundary of phonemes is determined by algorithm with maximum frequency in a short interval. Recognition process is performed by Continuous Hidden Markov Model(CHMM), and we compared it with another phoneme divided by eye-measurement. For the experiments result we confirmed that the method we presented is relatively superior in auto-segmentation in korean speech.

• Proceedings of the KSPS conference
• /
• 2005.11a
• /
• pp.159-162
• /
• 2005

Identification is the process automatically identify who is speaking on the basis of information obtained from speech waves. In training phase, each speaker models are trained using each speaker's speech data. GMMs (Gaussian Mixture Models), which have been successfully applied to speaker modeling in text-independent speaker identification, are not efficient in insufficient training data environment. This paper proposes speaker modeling method using MLLR (Maximum Likelihood Linear Regression) method which is used for speaker adaptation in speech recognition. We make SD-like model using MLLR adaptation method instead of speaker dependent model (SD). Proposed system outperforms the GMMs in small training data environment.

• Phonetics and Speech Sciences
• /
• v.10 no.4
• /
• pp.121-125
• /
• 2018

This paper describes procedures of changing and measuring voice quality in terms of jitter. Jitter synthesis method was applied to the TD-PSOLA analysis system of the Praat software. The jitter component is synthesized based on a Gaussian random noise model. The TD-PSOLA re-synthesize process is used to synthesize the modified voice with artificial jitter. Various vocal jitter parameters are used to measure the change in quality caused by artificial systematic jitter change. Synthetic vowels, natural vowels and short sentences are used to check the change in voice quality through the synthesizer model. The results shows that the suggested method is useful for voice quality control in a limited way and can be used to alter the jitter component of voice.

• The Bulletin of The Korean Astronomical Society
• /
• v.44 no.2
• /
• pp.59.2-59.2
• /
• 2019

In this talk, I will explain the implications of a rapid appearance of dark energy between the redshifts ($z$) of one and two on the expansion rate and growth of perturbations. Using both Gaussian process regression and a parametric model, I show that this is the preferred solution to the current set of low-redshift ($z<3$) distance measurements if $H_0=73~\rm km\,s^{-1}\,Mpc^{-1}$ to within 1\% and the high-redshift expansion history is unchanged from the $\Lambda$CDM inference by the Planck satellite. Dark energy was effectively non-existent around $z=2$, but its density is close to the $\Lambda$CDM model value today, with an equation of state greater than $-1$ at $z<0.5$. If sources of clustering other than matter are negligible, we show that this expansion history leads to slower growth of perturbations at $z<1$, compared to $\Lambda$CDM, that is measurable by upcoming surveys and can alleviate the $\sigma_8$ tension between the Planck CMB temperature and low-redshift probes of the large-scale structure.

• The Bulletin of The Korean Astronomical Society
• /
• v.45 no.1
• /
• pp.49.1-49.1
• /
• 2020

In this talk, I will discuss a few different techniques to reconstruct different cosmological functions, such as the primordial power spectrum and the expansion history. These model independent techniques are useful because they can discover surprising results in a way that nested modeling cannot. For instance, we can use the modified Richardson Lucy algorithm to reconstruct a novel primordial power spectra from the Planck data that can resolve the "Hubble tension". This novel primordial power spectrum has regular oscillatory features that would be difficult to find using parametric methods. Further, we can use Gaussian process regression to reconstruct the expansion history of the Universe from low-redshift distance datasets. We can also this technique to test if these datasets are consistent with one another, which essentially allows for this technique to serve as a systematics finder.

• Proceedings of the Korean Institute of Navigation and Port Research Conference
• /
• 2018.05a
• /
• pp.41-42
• /
• 2018

This study is about the prediction of ship's roll motion characteristic which has been used for evaluating ship's seakeeping performance. In order to obtain the ship's roll RAO during voyage, this paper utilized machine learning-based surrogate model. By comparing the prediction result data of surrogate model with test data, we suggest the best approximation technique and data sampling interval of the surrogate model appropriate for predicting the ships' roll motion characteristic.

• Journal of Korea Multimedia Society
• /
• v.19 no.8
• /
• pp.1587-1596
• /
• 2016

This paper presents an interactive locomotion controller using motion capture data and inverted pendulum model. Most of the data-driven character controller using motion capture data have two kinds of limitation. First, it needs many example motion capture data to generate realistic motion. Second, it is difficult to make natural-looking motion when characters navigate dynamic terrain. In this paper, we present a technique that uses dimension reduction technique to motion capture data together with the Gaussian process dynamical model (GPDM), and interpolates the low-dimensional data to make final motion. With the low-dimensional data, we can make realistic walking motion with few example motion capture data. In addition, we apply the inverted pendulum model (IPM) to calculate the root trajectory considering the real-time user input upon the dynamic terrain. Our method can be used in game, virtual training, and many real-time applications.