In this paper, we will examine a Korean large vocabulary continuous speech recognition (LVCSR) system for broadcast news speech. The combined vowel and implosive unit is included in a phone set together with other short phone units in order to obtain a longer unit acoustic model. The effect of this unit is compared with conventional phone units. The dictionary units for language processing are automatically extracted from eojeols appearing in transcriptions. Triphone models are used for acoustic modeling and a trigram model is used for language modeling. Among three major speaker groups in news broadcasts-anchors, journalists and people (those other than anchors or journalists, who are being interviewed), the speech of anchors and journalists, which has a lot of noise, was used for testing and recognition.

A psychoacoustic model based noise shaping method which shapes the noise in the frequency domain is proposed, where its presence with a host signal will not be perceptually noticeable. The derivation of imperceptible noise levels from the masking thresholds of the signal involves a deconvolution associated with the spreading function in the psychoacoustic model, which results in an ill-conditioned problem. In this paper, the problem is formulated as a constrained optimization, and it is demonstrated that the solution provides noise shaping where the noise excitation level conforms to the masking thresholds of the signal, and thus the noises embedded in the signal will not be perceived by human ear.

In this paper, we propose the bootstrap and aggregating (bagging) vector quantization (VQ) classifier to improve the performance of the text-independent speaker recognition system. This method generates multiple training data sets by resampling the original training data set, constructs the corresponding VQ classifiers, and then integrates the multiple VQ classifiers into a single classifier by voting. The bagging method has been proven to greatly improve the performance of unstable classifiers. Through two different experiments, this paper shows that the VQ classifier is unstable. In one of these experiments, the bias and variance of a VQ classifier are computed with a waveform database. The variance of the VQ classifier is compared with that of the classification and regression tree (CART) classifier[1]. The variance of the VQ classifier is shown to be as large as that of the CART classifier. The other experiment involves speaker recognition. The speaker recognition rates vary significantly by the minor changes in the training data set. The speaker recognition experiments involving a closed set, text-independent and speaker identification are performed with the TIMIT database to compare the performance of the bagging VQ classifier with that of the conventional VQ classifier. The bagging VQ classifier yields improved performance over the conventional VQ classifier. It also outperforms the conventional VQ classifier in small training data set problems.

In this paper, we propose a new method for selecting the threshold on wavelet packet denoising. In selecting threshold, the method using median is not efficient. Because this method can not recover unvoiced signal corrupted by noise. So we partition a speech signal corrupted by noise into the pure noise section and voiced section using autocorrelation and entropy. The autocorrelation and entropy can reflect disorder of noise. The new method yields more improved denoising effect. Especially unvoiced signal is very nicely reconstructed, and SNR is improved.

This study considers the environmental factors affecting propagation loss and sonar performance in the continental regions of the East Coast Sea of Korea. Water mass distributions appear to change dramatically in a few weeks. Simple calculation with the case when the NKCW (North Korean Cold Water) develops shows that the difference in propagation loss may reach in the worst up to 10dB over range 5km. Another factor, an eddy, has typical dimensions of 100-200km in diameter and 150-200m in thickness. Employing a typical eddy and assuming frequency to be 100Hz, its effects on propagation loss appear to make lower the normal formation of convergence zones with which sonars are possible to detect long-range targets. The change of convergence zones may result in 10dB difference in received signals in a given depth. Thermal fronts also appear to be critical restrictions to operating sonars in shallow waters. Assuming frequency to be 200Hz, thermal fronts can make 10dB difference in propagation loss between with and without them over range 20km. An observation made in one site in the East Coast Sea of Korea reveals that internal waves may appear in near-inertial period and their spectra may exist in periods 2-17min. A simulation employing simple internal wave packets gives that they break convergence zones on the bottom, causing the performance degradation of FOM as much as 4dB in frequency 1kHz. An acoustic experiment, using fixed source and receiver at the same site, shows that the received signals fluctuate tremendously with time reaching up to 6.5dB in frequencies 1kHz or less. Ambient noises give negative effects directly on sonar performance. Measurements at some sites in the East Coast Sea of Korea suggest that the noise levels greatly fluctuate with time, for example noon and early morning, mainly due to ship traffics. The average difference in a day may reach 10dB in frequency 200Hz. Another experiment using an array of hydrophones gives that the spectrum levels of ambient noises are highly directional, their difference being as large as 10dB with vertical or horizontal angles. This fact strongly implies that we should obtain in-situ information of noise levels to estimate reasonable sonar performance. As one of non-stationary noise sources, an eel may give serious problems to sonar operation on or under the sea bottoms. Observed eel noises in a pier of water depth 14m appear to have duration time of about 0.4 seconds and frequency ranges of 0.2-2.8kHz. The 'song'of an eel increases ambient noise levels to average 2.16dB in the frequencies concerned, being large enough to degrade detection performance of the sonars on or below sediments. An experiment using hydrophones in water and sediment gives that sensitivity drops of 3-4dB are expected for the hydrophones laid in sediment at frequencies of 0.5-1.5kHz. The SNR difference between in water and in sediment, however, shows large fluctuations rather than stable patterns with the source-receiver ranges.

The characteristics of radiation impedance are analyzed by algorithms which consists of Finite Element Method (FEM) and Hybrid type Infinite Element Method. The changes of radiation impedance for piston source according to the size and the material properties of baffles are studied. The results of the radiation impedance for rigid finite baffle coincide with other reports. The more the material properties of baffle that comes across the acoustic medium, the more the calculation results of radiation impedance approach the ones without baffle. Therefore, these results can be applied to the design and the radiation characteristics analysis of acoustic transducers.

Equations of motion of an engine mount system including foundation flexibility are derived. Forced vibration analysis is carried out for the given engine mount system excited with the unbalanced force and moment. A new optimal design method for the engine mount system is proposed, in which vibration characteristics of the chassis frame structure are considered as design parameters.