Search

Detection of the pitch frequency as the basic characteristic of the voice is one of the most important research subjects in voice analysis and synthesis. In this paper we study, in comparison with the cepstrum method and the modified correlation method, how the BPFP method using bandpass filter pairs and the BPFP-NN method combining the BPFP banks and the neural network (NN) proposed by the authors perform in the presence of pitch disturbance, waveform variation, and added noise when applied to voiced/unvoiced (U/V) detection and pitch extraction. According to the experimental results, the BPFP method and the BPFP-NN method exhibit more stable performance and more effective pitch extraction with U/V detection than the typical cepstrum method and modified correlation method, and the effect is particularly evident at low frequencies due to the logarithmic expression of the center frequency spacing of the BPFP banks. © 2003 Wiley Periodicals, Inc. Electron Comm Jpn Pt 3, 86(5): 24–35, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjc.10045 [ABSTRACT FROM AUTHOR]

Instead of the conventional static memory model based on an associative memory neural network, this paper proposes a dynamic memory model that is extended to the input-output system in order to allow interaction with external information. Based on the time correlation between the input signal and the network output, we analyze and investigate the network response to stored, nonstored, and mixed stored pattern signals, and the parameter dependencies inherent in the model. The possibility of a new mechanism, the chaotic dynamic model, which discriminates and identifies pattern signal information by combining the dynamic memory model with its extension is shown to be possible. © 2003 Wiley Periodicals, Inc. Electron Comm Jpn Pt 3, 86(5): 54–64, 2003; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjc.10022 [ABSTRACT FROM AUTHOR]

The Hopfield neural network is expected to act as the engine to solve the association problem or the optimization problem.In the associative processing using the Hopfield neural network, it is necessary the the pre-calculated weights should be memorized in the neural network. In the Hopfield neural network, however, there must be m² weights, where m is the number of neurons. In other words, a tremendous memory capacity is required when m is increased. The weight can be stored either by an analog memory or a digital memory. The analog memory, however, has problems of error and the decay of the memorized content. Then, digital memory must be used. In this case, the required number of memory elements is m² × (data bits) and the hardware complexity increases in proportion to the number of data bits. Another point is that the hardware, such as the multiplier of the synapse, becomes more complicated with the increase of the number of bits. A remedy for those problems may be to reduce the numbner of bits for the weight. When a digital code of some 2 bits is used to represent the weight, for example, the memory capacity for storage can be simplified. One of the important issues then is how to determine the weight when the number of bits for the weight is reduced. Since it is difficult to approximate the weight by a continuous variable, it is desirable to realize an efficient method which is different from the existing method of weight determination. With the foregoing as backgroun, this paper discusses the weight determination method for the case where the weight can take only a few number of discrete values. In the proposed method, the evaluation function for the weight is examined and the weights are determined by optimization using an algorithm, which is an application of the branch-and-bound method. It is shown by simulation that a high associative ability is realized by the association processing using the weighs obtained by the proposed method. Future applications can be expected. [ABSTRACT FROM AUTHOR]

We have been working toward the construction of a forward model arm employing a neural network model with electromyographic signals as the control input. We succeeded in estimating 1) the joint torque during isometric contraction in a plane from the electromyographic signals. 2) the path of motion from the degree of joint acceleration as well as 3) the path of motion from joint torque and 4) the position in three-dimensional space. In this paper, we present a new human interface employing a model of an arm, robot control of an artificial hand, and the learning of motion capability. In addition, based on the fact that we can estimate the position of the arm in three-dimensional space, we discuss as an example of the human interface the results of generating motion from electromyographic signals in a hypothetical arm using a neural network model. [ABSTRACT FROM AUTHOR]

Finding all solutions based on the piecewise-linear search is one of the steadiest methods to solve the non-linear equation. A feature of the method is that the local convergence problem is not involved. On the other hand, the number of region searches required even for a small-scale problem is tremendous. This paper proposes an efficient method for finding all solutions, restricting the problem to the analysis of bipolar circuits. Based on the range of values that each variable can take and the linearity of the equation, the solution candidate regions are restricted by evaluating the existence possibility of the solution for each of the multiple subset variables. The submatrix is considered and the size is varied, and the efficiency of the solution search is improved by testing a large number of linear regions at the same time. The proposed method works especially effectively when the signal propagation in the circuit has a directivity or the subcircuits have high independencies. Finally, the effectiveness of the proposed method is shown through actual computation examples. [ABSTRACT FROM AUTHOR]

When a random noise is superposed on a non-Gaussian signal with an abruptly changing component, as in the ease of the image signal, it is impossible for the conventional linear filter to eliminate the noise effectively. This is one of the cases where the nonlinear filter should be applied. This paper proposes a nonlinear digital filter based on the layered neural network, including the optimization of the nonlinear function. The proposed filter can effectively eliminate the noise by utilizing the nonlinearity of the layered neural network and its learning function. Another point is that the performance can be improved by the piecewise-linear approximation of the nonlinear function and the automatic design of the optimal nonlinear function. The effectiveness of the proposed filter is demonstrated by eliminating the noise from the two-dimensional image signal. [ABSTRACT FROM AUTHOR]

Neural networks have attracted attention due to their capability to perform nonlinear function approximation. In this paper, in order to better understand this capability, a new theorem on an integral transform was derived by applying ridge functions to neural networks. From the theorem, it is possible to obtain approximation bounds which clarify the quantitative relationship between the function approximation accuracy and the number of nodes in the hidden layer. The theorem indicates that the approximation accuracy depends on the smoothness of the target function. Furthermore, the theorem also shows that this type of approximation method differs from usual methods and is able to escape the so-called ‘curse of dimensionality,’ in which the approximation accuracy depends strongly of the input dimension of the function and deteriorates exponentially. [ABSTRACT FROM AUTHOR]

Linear adaptive digital filters are applied to various fields for their simplicity in the design and implementation. Considering many kinds of nonlinearities inherent to practical systems, however, nonlinear adaptive filtering will be more desirable. In this paper, we present a design method for parallel nonlinear adaptive digital filters using multilayer nonrecursive neural networks. Linear adaptive digital filters with the advantage of obtaining better accuracy of estimation. Furthermore, the nonlinear adaptive digital filter is extended to parallel nonlinear adaptive digital filters which reduce the computations of each filtering system, especially when a high-order nonlinear adaptive digital filter is necessary. [ABSTRACT FROM AUTHOR]

This paper presents a design method for the equal-ripple, linear-phase FIR digital filter when the tolerance for the error of magnitude response in the passband or stopband is specified. To realize the design when various constraints are imposed on the transfer function or magnitude response, the design problem is formulated as the optimization problem. A solution method is shown to solve the problem using the constrained neural net. When the lower bound of the magnitude error in the stopband is set as zero in the proposed method, the filter with double zeros in the stopband can easily be designed. Examples of application are shown for the minimum-phase FIR filter and the matched filter pair with zero intersymbol interference. [ABSTRACT FROM AUTHOR]

A novel adder and multiplier using Hopfield-type neural networks for which a technique is introduced to avoid the false operations is proposed, and the simulation results for these operations are illustrated. The main feature of these circuits is high-speed operation which is achieved at the almost constant computation time due to their parallel calculation ability even if the number of bits in digital circuits increases. The previously proposed adder has a problem in that some false operations occur in the special cases of addition. The error between the correct and false values in the operations is always only a binary value of ±1 once the false operation occurs. This paper describes in detail the technique to avoid the false operations. It is important to determine the ratio of weight of the object function to constraint function, taking into consideration the energy function that should be chosen. Computer simulations for the adder verify that the correct operation can be achieved by choosing the special ratio of the weight for these functions. On the other hand, a novel multiplier is described using a Hopfield-type network which is composed of the proposed adder and neural network to perform a logical product. The simulation results for the multiplier show that the correct operations of a multiplication can be achieved with high-speed performance. It is clear from these results that a high-performance digital adder and multiplier can be constructed by using the Hopfield-type neural network model. [ABSTRACT FROM AUTHOR]

Several methods have been presented recently for constructing the one-dimensional IIR digital filter by combining all-pass networks. It is known that the filter composed of the all-pass networks has a better computational efficiency and also low coefficient sensitivity due to the structural boundaries. In addition to the one-dimensional filters, a construction of the two-dimensional filter based on the all-pass network has also been proposed. A problem in that method, however, is the restriction on the realizable characteristics. This paper attempts to overcome that problem, and proposes a new construction of the two-dimensional digital filter using the all-pass network. The new approach employs the complex coefficient all-pass network instead of the conventional real-coefficient all-pass network. By this approach, the degree of freedom in the design is enhanced, and the characteristics which have been difficult or impossible to realize in the case of the real coefficient filters, are now made realizable. [ABSTRACT FROM AUTHOR]