An integrated pen interface system was developed to allow effective Japanese text entry. It consists of sub-systems for handwriting recognition, contextual post-processing, and enhanced Kana-to-Kanji conversion. The recognition sub-system uses a hybrid algorithm consisting of a pattern matcher and a neural network discriminator. Special care was taken to improve the recognition of non-Kanji and simple Kanji characters frequently used in fast data entry. The post-processor predicts consecutive characters on the basis of bigrams modified by the addition of parts of speech and substitution of macro characters for Kanji characters. A Kana-to Kanji conversion method designed for ease of use with a pen interface has also been integrated into the system. In an experiment in which 2,900 samples of Kanji and non-Kanji characters were obtained from 20 subjects, it was observed that the original recognition accuracy of 83.7% (the result obtained by using the pattern matching recognizer) was improved to 90.7% by adding the neural network discriminator, and that it was further improved to 94.4% by adding the post-processor. The improved recognition accuracy for non-Kanji characters was particularly marked.

We study robot navigation in unknown environment with rectangular obstacles aligned with the x and y axes. We propose a strategy called the modified-bian heuristic, and analyze its efficiency. Let n be the distance between the start point and the target of robot navigation, and let k be the maximum side length among the obstacles in a scene. We show that if k=(o(n) and if the summation of the widths of the obstacles on the line crossing the target and along the y axis is o(n), then ratio of the total distance walked by the robot to the shortest path length between the start point and the target is at most arbitrarily close to 1+k/2, as n grows. For the same restrictions as above on the sizes of the obstacles, the ratio is also at most arbitrarily close to 1+3/4n, as n grows, where is the summation of lengths of the obstacles in y axis direction.

A bounded complex (BC) digital transfer function realized with a II-cascade structure of Lossless Bounded Complex (LBC) two-pairs is known to have low magnitude sensitivity. In this letter, it is shown that the two-pairs parameters depend directly on some invariants of the transfer function corresponding to the transmission zeros of the structure. An analysis of the existence and the numbering of these invariants leads to a simplified automated LBC filter structure design avoiding the need for polynomial manipulations. These results are also easily applied for the real filtering case.

This paper shows that when a pattern matching method used in optical character readers is highly accurate, it can be used effectively in on-line Japanese character recognition. Stroke matching methods used in previous conventional on-line character recognition have restricted the number and the order of strokes. On the other hand, orientation-feature pattern matching methods avoid these restrictions. The authors have improved a pattern matching method with the development in the flexible pattern matching (FPM) method, based on nonlinear shape normalization and nonlinear pattern matching, which includes the normalization-cooperative feature extraction (NCFE) method. These improvements have increased the recognition rate from 81.9% to 95.9%, when applied to the off-line database ETL-9 from the Electrotechnical Laboratory, Japan. When applied on-line to the examination of 151,533 Kanji and Hiragana characters in 3,036 categories, the recognition rate achieved 94.0%, while the cumulative recognition rate within the best ten candidates was 99.1%.

Binary decision diagrams (BDD's) are graph representations of Boolean functions, and at the same time they can be regarded as a computational model. In this paper, we discuss relations between BDD's and other computational models and clarify the computational power of BDD's. BDD's have the property that each variable is examined only once according to a total order of the variables. We characterize families of BDD's by on-line deterministic Turing machines and families of permutations. To clarify the computational power of BDD's, we discuss the difference of the computational power with respect to the way of reading inputs. We also show that the language TADGAP (Topologically Arranged Deterministic Graph Accessibility Problem) is simultaneously complete for both of the class U-PolyBDD of languages accepted by uniform families of polynomial-size BDD's and the clas DL of languages accepted by log-space bounded deterministic Turing machines. From the results, we can see that the problem whether U-PolyBDD U-NC1 is equivalent to a famous open problem whether DL U-NC1, where U-NC1 is the class of languages accepted by uniform families of log-depth constant fan-in logic circuits.

We consider a class of unknown scenes Sk(n) with rectangular obstacles aligned with the axes such that Euclidean distance between the start point and the target is n, and any side length of each obstacle is at most k. We propose a strategy called the adaptive-bias heuristic for navigating a robot in such a scene, and analyze its efficiency. We show that a ratio of the total distance walked by a robot using the strategy to the shortest path distance between the start point and the target is at most 1+(3/5) k, if k=o(n) and if the start point and the target are at the same horizontal level. This ratio is better than a ratio obtained by any strategy previously known in the class of scenes, Sk(n), such that k=o(n).

This paper presents a new method for resolving lexical (word sense) ambiguities inherent in natural language sentences. The Sentence Analyzer (SENA) was developed to resolve such ambiguities by using constraints and example-based preferences. The ambiguities are packed into a single dependency structure, and grammatical and lexical constraints are applied to it in order to reduce the degree of ambiguity. The application of constraints is realized by a very effective constraint-satisfaction technique. Remaining ambiguities are resolved by the use of preferences calculated from an example-base, which is a set of fully parsed word-to-word dependencies acquired semi-automatically from on-line dictionaries.

The requirement of structural boundedness or passivity leads to important classes of digital filters among which are the wave digital (WD) filters and the LBR cascade structures having low coefficient sensitivity. Contrary to the WD filters, the LBR filters are directly synthesized in z-domain and several authors presented different approaches for a better understanding of the synthesis procedure especially for complex transfer functions. Some tentatives were also made to give parallels between passive analog and digital filters (i.e. WD or LBR filters). A general approach to LBR synthesis with transmission zeros not necessarily on the unit circle is presented along with some explicit expressions for the LBR (and the generalized complex counterpart LBC) filter parameters for the realization of an input transfer function. The results can be of interest in automated procedures for low sensitivity digital filter design.

For the production of high quality synthetic sounds in a text-to-speech system, an excellent synthesizing method of speech signals is indispensable. In this paper, a new speech analysis-synthesis method for the text-to-speech system is proposed. The signals of voiced speech, which have a line spectrum structure at intervals of pitch in the linear frequency domain, can be represented approximately by the superposition of sinusoidal waves. In our system, analysis and synthesis are performed using such a harmonic structure of the signals of voiced speech. In the analysis phase, assuming an exact harmonic structure model at intervals of pitch against the fine structure of the short-time power spectrum, the fundamental frequency f0 is decided so as to minimize the error of the log-power spectrum at each peak position. At the same time, according to the value of the above minimized error, the rate of periodicity of the speech signal is detemined. Then the log-power spectrum envelope is represented by the cosine-series interpolating the data which are sampled at every pitch period. In the synthesis phase, numerical solutions of non-linear differential equations which generate sinusoidal waves are used. For voiced sounds, those equations behave as a group of mutually synchronized oscillators. These sinusoidal waves are superposed so as to reconstruct the line spectrum structure. For voiceless sounds, those non-linear differential equations work as passive filters with input noise sources. Our system has some characteristics as follows. (1) Voiced and voiceless sounds can be treated in a same framowork. (2) Since the phase and the power information of each sinusoidal wave can be easily controlled, if necessary, periodic waveforms in the voiced sounds can be precisely reproduced in the time domain. (3) The fundamental frequency f0 and phoneme duration can be easily changed without much degradation of original sound quality.

A least squares approximation method of recursive digital filters for finite interval response with zero value outside the interval is presented. According to the characteristic of the method, the modified Gauss Method is utilized in iteratively determining design parameters. Convergence, together with the stability of the resulting filter, are guaranteed.

A systematic method is proposed to generate a random image with a known correlation function and the modified Laplace distribution; the modified Laplace distribution includes the one-side exponential distribution and the Laplace distribution as a special case. Several random images with an isotropic correlation and the modified Laplace distribution are generated and displayed in figures.

Asynchronous machines are a topic of great interest in the research area of actuators. Due to the complexity of these systems and to the required performance, the modelling and control of asynchronous machines are complex questions. Problems arise when the control goals require accurate descriptions of the electric machine or when we need to identify some electrical parameters; in the models employed it becomes very hard to take into account all the phenomena involved and therefore to make the error amplitude adequately small. Moreover, it is well known that, though an efficient control strategy requires knowledge of the flux vector, direct measurement of this quantity, using ad hoc transducers, does not represent a suitable approach, because it results in expensive machines. It is therefore necessary to perform an estimation of this vector, based on adequate dynamic non-linear models. Several different strategies have been proposed in literature to solve the items in a suitable manner. In this paper the authors propose a neural approach both to derive NARMAX models for asynchronous machines and to design non-linear observers: the need to use complex models that may be inefficient for control aims is therefore avoided. The results obtained with the strategy proposed were compared with simulations obtained by considering a classical fifth-order non-linear model.

This paper proposes a superposed SSMA (Spread Spectrum Multiple Access)-QPSK (Quadrature Phase Shift Keying) signal transmission scheme over high speed QPSK signals to achieve higher frequency utilization efficiency and to facilitate lower power transmitters for SSMA-QPSK signal transmission. Experimental results show that the proposed scheme which employs the coding-rate of one-half FEC (Forward Error Correction) and a newly proposed co-channel interference cancellation scheme for SSMA-QPSK signals can transmit twenty SSMA-QPSK channels simultaneously over a nonlinearly amplified high speed QPSK signal transmission channel and achieve as ten times SSMA channels transmission as that without co-channel interference cancellation when the SSMA-QPSK signal power to the high speed QPSK signal power ratio equals -30dB. Moreover, cancellation feasibility generation of the interference signals replica through practical hardware implementation is clarified.

This paper describes a new trial of dynamical parameter estimation for the actual room acoustic system, in a practical case when the input excitation is polluted by a background noise in contrast with the usual case when the output observation is polluted. The room acoustic system is first formulated as a discrete time model, by taking into consideration the original standpoint defining the system parameter and the existence of the background noise polluting the input excitation. Then, the recurrence estimation algorithm on a reverberation time of room is dynamically derived from Bayesian viewpoint (based on the statistical information of background noise and instantaneously observed data), which is applicable to the actual situation with the non-Gaussian type sound fluctuation, the non-linear observation, and the input background noise. Finally, the theoretical result is experimentally confirmed by applying it to the actual estimation problem of a reverberation time.

Information geometry is a new powerful method of information sciences. Information geometry is applied to manifolds of neural networks of various architectures. Here is proposed a new theoretical approach to the manifold consisting of feedforward neural networks, the manifold of Boltzmann machines and the manifold of neural networks of recurrent connections. This opens a new direction of studies on a family of neural networks, not a study of behaviors of single neural networks.