The naturalness of normal sustained vowels is considered to be attributable to the fluctuations observed in the steady part where speech signal is seemingly almost periodic. There always exist two kinds of involuntary fluctuations in the steady part of sustained vowels, even if the sustained vowels are phonated as steadily as possible. One is pitch period fluctuation and the other is waveform fluctuation. In this study, frequency analyses on these fluctuations were conducted in order to investigate their general characteristics. The results of the analyses suggested that the frequency characteristics of the fluctuations were possible to be approximated as 1/fβ-like, which is regarded as the specific feature of random fractal. Therefore, a procedure based on random fractal generation methods was proposed in order to produce these fluctuations for the improvement of the voice quality of synthesized sustained vowels. A series of psychoacoustic experiments was also conducted to evaluate the proposed technique. Experimental results indicated that the proposed technique was effective for synthesized sustained vowels to be perceived as human-like. Unlike the sustained vowels which were synthesized without pitch period fluctuation nor waveform fluctuation, the synthesized sustained vowels which contained the fluctuations were not perceived as buzzer-like, which is the major problem of the voice quality of synthesized sustained vowels. However, it was also found that both of the fluctuations were not always the acoustic cues for the naturalness of normal sustained vowels. The synthesized sustained vowels which contained the fluctuations whose frequency characteristics were the same as that of white noise were perceived as noise-like, which is not at all the voice quality of normal sustained vowels. The results of psychoacoustic experiments indicated that the frequency characteristics of the fluctuations, which are possible to be modeled as 1/fβ-like, were the significant factors for the naturalness of normal sustained vowels.

In order to improve microprocessor performance, we propose to utilize histories of dynamic instruction sequences. A lot of special purpose memories integrated in a processor chip hold the histories. In this paper, we describe the usefulness of using two special purpose memories: Non-Consecutive basic block Buffer (NCB) and Reference Prediction Table (RPT). The NCB improves instruction fetching efficiency in order to relieve control dependences. The RPT predicts data addresses in order to speculate data dependences. From the simulation study, it has been found that the proposed mechanisms improve processor performance by up to 49. 2%.

We propose an information integration architecture for a man-machine interface to construct a new agent-based Computer Supported Cooperative Work (CSCW) system. The system acts as a clerk in cooperative work giving users the advantage of using cooperative work space. The system allows users to do their work in the style of an ordinary meeting because spontaneous expressions of speech and gestures by users are detected by sensors so that they can be integrated with a task model at several levels to create suitable responses in a man-machine interface. As a result, users can dedicate themselves to mutually understand other meeting members with no awareness of direction to the CSCW system. In this paper, we describe the whole system and its information integration architecture for the man-machine interface including, the principle of functions, the current status of the system and future directions.

We fabricated a new lateral Power MOS FET with an SOI structure formed by a solid phase epitaxy method which has a buried gate under the channel layer and a trench gate/drain. We studied the temperature dependence of the Vg-Id characteristic and Vd-Id characteristic in the temperature range of from 27 to 150. In the Vd-Id characteristic, the drain current decreases as the temperature increases in the saturation region, but the drain current increases as the temperature increases in the linear region. Zero-temperature-coefficient bias point of Vg-Id characteristic was 0. 3 V. And, the threshold voltage variation was -2. 8 mV/. In the Vd-Id characteristic, the drain current decreases as the temperature increases. The specific on-resistance was obtained in the linear region of the Vd-Id characteristic. The specific on-resistance variation was 0. 3 mWmm2/. The temperature characteristics of this device are as good as those of the conventional MOS FET.

Conventional approaches to neural network training do not consider possibility of selecting training samples dynamically during the learning phase. Neural network is simply presented with the complete training set at each iteration of the learning. The learning can then become very costly for large data sets. Huge redundancy of data samples may lead to the ill-conditioned training problem. Ill-conditioning during the training causes rank-deficiencies of error and Jacobean matrices, which results in slower convergence speed, or in the worst case, the failure of the algorithm to progress. Rank-deficiencies of essential matrices can be avoided by an appropriate selection of training exemplars at each iteration of training. This article presents underlying theoretical grounds for dynamic sample selection (DSS), that is mechanism enabling to select a subset of training set at each iteration. Theoretical material is first presented for general objective functions, and then for the objective functions satisfying the Lipschitz continuity condition. Furthermore, implementation specifics of DSS to first order line search techniques are theoretically described.

Computational expensiveness of the training techniques, due to the extensiveness of the data set, is among the most important factors in machine learning and neural networks. Oversized data set may cause rank-deficiencies of Jacobean matrix which plays essential role in training techniques. Then the training becomes not only computationally expensive but also ineffective. In [1] the authors introduced the theoretical grounds for dynamic sample selection having a potential of eliminating rank-deficiencies. This study addresses the implementation issues of the dynamic sample selection based on the theoretical material presented in [1]. The authors propose a sample selection algorithm implementable into an arbitrary optimization technique. An ability of the algorithm to select a proper set of samples at each iteration of the training has been observed to be very beneficial as indicated by several experiments. Recently proposed approaches to sample selection work reasonably well if pattern-weight ratio is close to 1. Small improvements can be detected also at the values of the pattern-weight ratio equal to 2 or 3. The dynamic sample selection approach, presented in this article, can increase the convergence speed of first order optimization techniques, used for training MLP networks, even at the value of the pattern-weight ratio (E-FP) as high as 15 and possibly even more.

Evaluating analytically computer architecture performance is mostly cheap and quick. However, existing analytical performance evaluation techniques usually have a difficult and time-consuming modeling process. Moreover, existing techniques do not support well the capability for finding the bottleneck and its cause of a target system being evaluated. To address the above problems and to enhance analytical performance evaluation technology, in this paper we propose a software tool that accepts system models described in a specification language, generating an executable program that performs the actual performance evaluation. The whole approach is built on a subsystem-oriented performance evaluation tool, which is, in turn, based on a formal subsystem-oriented performance evaluation technique and a subsystem specification language.

Recent progress in neural network research has demonstrated the usefulness of neural networks in a variety of areas. In this work, its application in the spatial resolution improvement of a remotely sensed low resolution thermal infrared image using high spatial resolution of visible and near-infrared images from Landsat TM sensor is described. The same work is done by an algebraic method. The tests developed are explained and examples of the results obtained in each test are shown and compared with each other. The error analysis is also carried out. Future improvements of these methods are evaluated.

The Heart Rate Variability (HRV) analysis has become vigorous these days. One reason for this is that the HRV analysis investigates the dynamics of the autonomic nervous system activities which control the HRV. The Integral Pulse Frequency Modulation (IPFM) model is a pulse generating mechanism model in the nervous system, that is one of the models which connects the HRV to the autonomic nervous system activities. The IPFM model is a single frequency component model; however, the real HRV has multiple frequency components. Moreover, there are refractory periods after generating action potentials are initiated. Nevertheless, the IPFM model does not consider refractory periods. In order to make sure of the accuracy and the effectiveness of the integral function (IF) method applied to the real data, we consider the absolute refractory periods and two frequency components. In this investigation, the simulated HRV was made with a single and double frequency component using the IPFM model with and without absolute refractory periods. The original generating function of the IPFM model was demodulated by using the instantaneous heart rate tachogram. The power of the instantaneous pulse rate per minute was analyzed by the direct FFT method, the IF FFT method without the absolute refractory periods, and the IF FFT method with the absolute refractory periods. It was concluded that the IF FFT method can demodulate the original generating function accurately.

To improve speech coding quality, in particular, the long-term dependency prediction characteristics, we propose a new nonlinear predictor, i. e. , a fully connected recurrent neural network (FCRNN) where the hidden units have feedbacks not only from themselves but also from the output unit. The comparison of the capabilities of the FCRNN with conventional predictors shows that the former has less prediction error than the latter. We apply this FCRNN instead of the previously proposed recurrent neural networks in the code-excited predictive speech coding system (i. e. , CELP) and shows that our system (FCRNN) requires less bit rate/frame and improves the performance for speech coding.

A VLSI implementation of a low-power DSP core is described, which is dedicated to the G. 723. 1 low bitrate speech codec. A number of sophisticated DSP microarchitectures are devised mainly on dual multiply accumulators, rounding and saturation mechanisms, and two-banked on-chip memory. The main attempt is focused on lowering the clock frequency, and therefore on reducing the total power consumption, at the cost of a fairly small increase of chip area. The proposed DSP architecture has been integrated in the total area of 7. 75 mm2 by using a 0. 35 µm CMOS technology, which can operate at 10 MHz with the dissipation of 44. 9 mW from a single 3 V supply.

Macroscopic method for quantization of the evanescent fields brought about by total reflection is presented. Here, a semi-infinite space is assumed to be filled with a transparent dispersive dielectric with dielectric constant ε(ω) to the left of the plane z = 0, and be empty to the right of the plane. The wave is assumed to be incident from the left, and so the whole field is composed of the triplet of incident, reflected, and transmitted waves labeled by a continuous wave vector index. The transmitted wave in free space may be evanescent. The triplet is shown exactly without using slowly varying field approximation in dispersive medium to form orthogonal mode for different wave vectors, which provides the basis for the quantization of the triplet with taken into account of medium dispersion. The exact orthogonal relation reduces to the well known one if the dielectric is nondispersive, ε/ω = 0. By using the field expansion in terms of the orthogonal triplet modes, the total field energy is found to be the sum of the energies of independent harmonic oscillators. A discussion is also made on the wave momentum of evanescent field.

Dense WDM techniques that exploit the enormous bandwidth of dispersion-shifted fibers (DSFs) while avoiding the impairments due to nonlinear effects are described. First, the nature of four-wave mixing (FWM), the dominant impairment factor in WDM transmission systems, is investigated using DSF installed in the field and laboratory experiments. This provides useful information for the practical design of WDM networks based on DSF. Second, practical techniques to reduce FWM impairment, unequal channel allocation and off-lambda-zero channel allocation (equal channel allocation in the novel 1580 nm band) along with gain-shifted erbium-doped fiber amplifiers for the 1570 to 1600 nm band, is described. Comparisons between off-lambda-zero and unequal channel allocation are provided in terms of the maximum transmission distance for various numbers of channels. Two schemes to immunize WDM systems against group velocity dispersion, span-by-span dispersion compensation and optical duobinary format, are presented. The combination of unequal channel allocation with off-lambda-zero channel allocation as well as the combination of two bands: the conventional 1550 nm band and the novel 1580 nm band are proven to be very useful in expanding the usable bandwidth of DSFs.

It becomes essential in practice to improve a processing rate and to divide an image into small segments adjusting a limited memory, because image filing systems handle large images up to A1 size. This paper proposes a new method of an automatic skew normalization, comprising a high-speed skew detection and a distortion-free dividing rotation. We have evaluated the proposed method from the viewpoints of the processing rate and the accuracy for typed documents. As results, the processing rate is 2. 9 times faster than that of a conventional method. A practical processing rate for A1 size documents can be achieved under the condition that the accuracy of a normalized angle is controlled within 0. 3 degrees. Especially, the rotation with dividing can have no error angle, even when the A1 size documents is divided into 200 segments, whereas the conventional method cause the error angle of 1. 68 degrees.

This paper describes the wavelength division multiplexing technology for the long-haul optical communication system, especially for the undersea cable system. At first, the present WDM technology for the undersea cable system is reviewed briefly. After that, some experiments using compensation of the dispersion slope of the transmission fiber are discussed as future technical options of undersea systems with over 100 Gbit/s capacity.

We have developed a multiple quantum well (MQW) electroabsorption (EA) modulator for wavelength-division multiplexing (WDM) switching systems. The fabricated MQW EA gate has low polarization and wavelength-dependent loss and high extinction ratio within the wavelength range of 1545 to 1560 nm. And by using this gate ultra-high-speed switching is achieved for WDM signals. Moreover, we optimize the EA gate for the full gain-band of an erbium-doped fiber amplifier (EDFA)(1535 to 1560 nm). This EA gate provides low polarization-dependent loss, higher extinction ratio, and high saturation input power in the wider wavelength range. These MQW EA gates will play an important role in future WDM switching systems.

The effectiveness of periodic dispersion compensation on single-channel 40 Gbit/s soliton transmission system was experimentally investigated. This technique requires just the dispersion compensation fibers and wideband optical filters in the transmission line, which has no difficulty to be used in the practical system. By using polarization-division-multiplexing together with periodic dispersion compensation, single-channel 40 Gbit/s transmission over 4700 km was demonstrated. Single-polarization 40 Gbit/s transmission experiments, which are more suitable for system implementation and compatible with WDM were also conducted. We investigated the transmission characteristics and pulse dynamics in different dispersion maps and in the optimized dispersion map, single-channel, single-polarization 40 Gbit/s transmission over 6300 km was successfully demonstrated.

An optimum filter for extracting a time-varying harmonic signal from the noise-corrupted measurement is proposed. It is derived as a solution of the least mean square estimation with consideration of the pitch estimation error even without any assumption on the filter model. We obtain a comb-like impulse response which consists of homologous and dilated distribution of weights just located periodically with a pitch interval. This remarkable structure is well suited to the proportionally expanding error of pitch repetition times. Examples of the filter design are presented, and the performance of noise suppression is examined by comparison with conventional comb filters.

This paper describes a classification method for rotated and scaled textured images using invariant parameters based on spectral-moments. Although it is well known that rotation invariants can be derived from moments of grey-level images, the use is limited to binary images because of its computational unstableness. In order to overcome this drawback, we use power spectrum instead of the grey levels to compute moments and adjust the integral region of moment evaluation to the change of scale. Rotation and scale invariants are obtained as the ratios of the different rotation invariants on the basis of a spectral-moment property with respect to scale. The effectiveness of the approach is illustrated through experiments on natural textures from the Brodatz album. In addition, the stability of the invariants with respect to the change of scale is discussed theoretically and confirmed experimentally.

This paper presents a novel transmission diversity scheme for code division multiple access system. Conventional diversity receivers in mobile stations require space and complicated circuits, however, the proposed diversity schemes present significant diversity effect without any diversity equipment at the mobile station. It is possible to use the transmitter diversity at the base station by using the feature of time division duplex (TDD) which has strongly correlated fading patterns in both forward and reverse link. Computer simulation is performed to evaluate the performance of the proposed systems for single user environment. The performance of the system 1, which select best situated antenna, is analyzed and the BER performance for multiple access is presented.