In this study, we examine the applicability of an artificial neural network(ANN) for filtering underwater random noise and for identifying underlying signals taken from noisy environment. The approach is to find a way of compressing the input data and then decompressing it using an ANN as in image compressing process. It is well known that random signal is hard to compress while ordered information is not. The use of a limited number of processing elements(PEs) in the hidden layer of an Ann ensures that some of the noise would be removed in the reconstruction process. Two types of the signals, synthesized and measured, are used to examine the effectiveness of the ANN-based filter. After training process is completed, the ANN successfully extracts the underlying signals form the synthesized or measured noisy signals. In particular, compared with the results form without filtering or moving averaged, the ANN-based filter gives much better spectrograms to identify underlying signals from the measured noisy data. This filtering process is achieved without using and kind of highly accurate signal processing technique. More experimentation needs to be followed to develop the ANN-based filtering technique to the level of complete understanding.

A real-time identifier for the nonstationary time-varying signals and systems was implemented using a low cost DSP (digital signal processing) chip. The identifier is comprised of I/O units, a central processing unit, a control unit and its supporting software. In order t estimate the system accurately and to reduce quantization error during arithmetic operation, the firmware was programmed with 64-bit extended precision arithmetic. The performance of the identifier was verified by comparing with the simulation results. The implemented real-time identifier has negligible quantization errors and its real-time processing capability crresponds to 0.6kHz for the nonstationary AR (autoregressive) model with n=4 and m=1.

In this paper, convergence properties of the LMS, LMF, and LVCMS algorithms are investigated under the assumption of the uncorrelated Gaussian input data. By treating these algorithms as special cases of more general algorithm family, unified results on these algorithms are obtained. First the upper bound on the step size parameter is obtained. Second, an expression for misadjustment is obtained. These theoretical results confirm earlier LMS works. Further, the results explain why the LMS and LVCMS algorithms are experiencing difficulties with plant noise having heavier tailed densities. Simulation results agree with theoretical expectation closely for various plant noise statistics.

A recursive estimation for the enhancement of white noise contaminated speech is proposed. This method is based on the Kalman filter with time-varying parametric model for the clean speech signal. Then, hidden filter model are used to model the clean speech signal. An approximation improvement of 4-5 dB in SNR is achieved at 5 and 10 dB input SNR, respectively.

In this paper, a two-stage recognition approach based on error pattern hypotheses is proposed to reduce errors of a connected digit recognizer. In the approach, a conventional recognizer is first used to produce N-best candidate strings, and then error patterns are hypothesized by examining the candidate strings. For substitution error pattern hypotheses, error-pattern-dependent classifiers having more discriminative power than the first-stage classifier are used ; and for insertion and deletion errors, word duration and energy contour information are exploited are exploited to discriminated confusing pairs. Simulation results showed that the proposed approach achieves 15% decrease in word error rate for speaker-independent Korean connected digit recognition when a hidden Markov model-based recognizer is used for the first-stage classifier.

A new method for efficient modeling of the Head-Related Transfer Functions(HRTF's) without loss of any directional information is proposed. In this paper, the HRTF's were empirically measured in a real room and modeled as the ARMA models with common AR coefficients and different MA coefficients. To assess the validity of the proposed ARMA model, psychophysical tests show that the proposed ARMA model, in comparison with the conventional MA model, requires a small number of parameters to represent empirical HRTF's and improves the back-to-front confusions in sound-field localization. Thus, significant simplifications in the implementations of sound-field synthesis systems could be obtained by using the proposed ARMA model.

Generalized iterative methods for reducing noise in contaminated chaotic signals are proposed. These methods minimize a cost function composed of two parts : one containing information that represents how close enhanced signals are to the observed signal and another composed of constraints that fit the dynamics of the system. The convergence conditions and the error systems of the proposed are investigated. As one aspect of noise reduction, the suppression or cancellation of a chaotic interference signal is discussed.

This paper describes on the Korean continuous speech recognition system for flight reservation. It adopts a frame-synchronous One-Pass DP search algorithm driven by syntactic constraints of context free grammar(CFG). For recognition, 48 phoneme-like units(PLU) were defined and used as basic units for acoustic modeling of Korean. This modeling was conducted using a HMM technique, where each model has 4-states 3-continuous output probability distributions and 3-discrete-duration distributions. Language modeling by CFG was also applied to the task domain of flight reservation, which consisted of 346 words and 422 rewriting rules. In the tests, the sentence recognition rate of 62.6% was obtained after speaker adaptation.

In this paper, we propose the filtered-x least mean fourth (LMF) algorithm where the error raised to the power of four is minimized and analyze its convergence behavior for a multiple sinusoidal acoustic noise and Gaussian measurement noise. Application of the filtered-x LMF adaptive filter to active noise control(ANC) requires estimating of the transfer characteristic of the acoustic path between the output and error signal of the adaptive controller. The results of the convergence analysis of the filtered-x LMF algorithm indicates that the effects of the parameter estimation inaccuracy on the convergence behavior of the algorithm are characterized by two distinct components : Phase estimation error and estimated gain. In particular, the convergence is shown to be strongly affected by the accuracy of the phase response estimate. Also, we newly show that convergence behavior can differ depending on the relative sizes of the Gaussian measurement noise and convergence constant.

In this paper we implement the Discrete Rotational Fourier Transform(DRFT) which is a discrete version of the Angular Fourier Transform and its inverse transform. We simplify the computation algorithm in [4], and calculate the complexity of the proposed implementation of the DRFT and the inverse DRFT, in comparison with the complexity of a DFT (Discrete Fourier Transform).

The set of admissible time-variations in the input signal can be separated into two categories : slow parameter changes and large parameter changes which occur infrequently. A common approach used in the tracking of slowly time-varying parameters is the exponential recursive least-squares(RLS) algorithm. There have been a variety of research works on the error analysis of the exponential RLS algorithm for the slowly time-varying parameters. In this paper, the focus has been given to the error analysis of exponential RLS algorithms for the input data with abrupt property changes. The voiced speech signal is chosen as the principal application. In order to analyze the error performance of the exponential RLS algorithm, deterministic properties of the exponential RLS algorithms is first analyzed for the case of abrupt parameter changes, the impulsive input(or error variance) synchronous to the abrupt change of parameter vectors actually enhances the convergence of the exponential RLS algorithm. The analysis has also been verified through simulations on the synthetic speech signal.

The aims of this study were to describe the way in which sound durations are perceived, and to attempt to explain the hidden mechanisms of the duration perception in music performances. Three experiments were carried out to determine the difference limen for the perception of sound duration and to find the effects of frequency and intensity on duration discrimination. For short duration tones ranging from 25 to 100 msec, a linear improvement in discrimination judgements was found with increasing duration of signal. The JND was constant for durations between 100 msec and 2 sec. However, for extended stimulus durations (more than 2 sec) the JND was again linearly improved. Subjects were also presented with a pair of stimuli, composed of high and low frequency pure tones, and asked to discriminate differences in duration of the two tones and ignore differences in the frequency of the tones. It was found that subjects perceived the higher frequency to be longer in duration. When an experiment was carried out to investigate the effect of intensity on duration discrimination it was found that a 20 phon difference makes subjects perceive the louder stimulus as longer than the quieter stimulus. Finally, in a performance study, an analysis of musical performances revealed the effect of frequency. It was found that the musicians played the higher notes shorter than the lower notes. This agrees with what was previously found in the work on the perception of tones.