A technique for reducing the footprints of HMM-based speech synthesis systems by tying all covariance matrices of state distributions is described. HMM-based speech synthesis systems usually leave smaller footprints than unit-selection synthesis systems because they store statistics rather than speech waveforms. However, further reduction is essential to put them on embedded devices, which have limited memory. In accordance with the empirical knowledge that covariance matrices have a smaller impact on the quality of synthesized speech than mean vectors, we propose a technique for clustering mean vectors while tying all covariance matrices. Subjective listening test results showed that the proposed technique can shrink the footprints of an HMM-based speech synthesis system while retaining the quality of the synthesized speech.

The "geometric AIC" and the "geometric MDL" have been proposed as model selection criteria for geometric fitting problems. These correspond to Akaike's "AIC" and Rissanen's "BIC" well known in the statistical estimation framework. Another well known criterion is Schwarz' "BIC", but its counterpart for geometric fitting has not been known. This paper introduces the corresponding criterion, which we call the "geometric BIC", and shows that it is of the same form as the geometric MDL. Our result gives a justification to the geometric MDL from the Bayesian principle.

Asymmetric Slope Dual Mode Differential Logic (ASDMDL) embodies high-speed dynamic and low-power static operations in a single design. Two-phase dual-rail logic signaling is used in a high-speed operation, where a logical evaluation is preceded by pre-charge, and it asserts one of the rails with an asymmetrically shortened rise transition to express a binary result. On the other hand, single-phase differential logic signaling eliminates pre-charge and leads to a low-power static operation. The operation mode is defined by the logic signaling styles, and no control signal is needed in the logic cell. The design of mixed CMOS and ASDMDL logic circuits can be automated with general logic synthesis and place-and-route techniques, since the physical ASDMDL cell is prepared in such a way to comply with a CMOS standard-cell design flow. A mixed ASDMDL/CMOS micro-processor in a 0.18-µm CMOS technology demonstrated 232 MHz operation, corresponding to 14% speed improvement over a full CMOS implementation. This was achieved by substituting ASDMDL cells for only 4% of the CMOS logic cells in data paths. The low-speed operation of ASDMDL at 100 MHz was nearly equivalent to that of CMOS. However, power consumption was reduced by 3% due to the use of ASDMDL complex logic cells. Area overhead was less than 4%.

This paper presents a scheme for accurately detecting the number of incident waves arriving at array antennas. The array antenna and MIMO techniques are developing as 4th generation mobile communication systems and wireless LAN technologies, and the accurate estimation of the propagation environment is becoming more important. This paper emphasizes the accurate detection of the number of incident waves; one of the important characteristics in multidirectional communication. There are some recent papers on accurate detection but they have problems of estimation accuracy or computational cost in severe environment like low SNR, small number of snapshots or waves with close angles. Hence, AIC and MDL methods based on statistics and information theory are still often used. In this paper, we propose an accurate estimation method of the number of arrival signals using the orthogonality of subspaces derived from preliminary estimation of signal subspace. The proposed method accurately estimates the number of signals also in severe environments where AIC and MDL methods can hardly estimate. We evaluate the performance of these methods through some computer simulation and experiments in anechoic chamber.

Multipath search based instantaneous root-mean-squared (RMS) delay spread (RDS) estimators mainly depend on path detection or multipath search. This paper proposes a novel method for multipath search through Minimum Descriptive Length (MDL) criterion, and hence a novel instantaneous RDS estimation method for wireless OFDM systems. compared with the conventional multipath search based instantaneous RDS estimators, the proposed estimator doesn't need any a priori information about the noise variance and the channel power delay profile (PDP) while the performance is improved. Simulation results demonstrate that the proposed estimator is also insensitive to the variance of SNR and robust against the frequency selectivity, as well as the vehicle speed.

We propose a new method to introduce the Minimum Description Length (MDL) criterion to the automatic generation of non-uniform, context-dependent HMM topologies. Phonetic decision tree clustering is widely used, based on the Maximum Likelihood (ML) criterion, and only creates contextual variations. However, the ML criterion needs to predetermine control parameters, such as the total number of states, empirically for use as stop criteria. Information criteria have been applied to solve this problem for decision tree clustering. However, decision tree clustering cannot create topologies with various state lengths automatically. Therefore, we propose a method that applies the MDL criterion as split and stop criteria to the Successive State Splitting (SSS) algorithm as a means of generating contextual and temporal variations. This proposed method, the MDL-SSS algorithm, can automatically create adequate topologies without such predetermined parameters. Experimental results for travel arrangement dialogs and lecture speech show that the MDL-SSS can automatically stop splitting and obtain more appropriate HMM topologies than the original one.

A novel adaptive technique based on a statistical model by estimating window size for unsupervised segmentation of a set of MR images is presented. The window size estimation is achieved in the image using a MDL for mixture estimation and segmentation, and allows the technique to well reflect local characteristics of the image.

Many experimentally and theoretically based models have been proposed to predict, quantitatively evaluate, and combat the fading phenomenon in mobile communication systems. However, to the best of the authors' knowledge, up to now there is no objective method to determine which is the most suitable distribution to model the fading phenomenon based on experimental data. In this work, the Minimum Description Length (MDL) criterion for model selection is proposed for that purpose. Furthermore, the MDL analysis is performed for some of the most widely used fading models based on measurements taken in a sub-urban environment.

The Matrix-Pencil (MP) method is applied to the estimation of the undesired radiation from the microstrip line discontinuities. The Q factors are obtained from the complex resonant frequencies estimated from FDTD transient field by using MP. The number of the damped oscillations is estimated by using MDL which is widely used as an information theoretic criterion for the model order estimation.

This paper addresses the problem of learning Bayesian belief networks (BBN) based on the minimum description length (MDL) principle. First, we give a formula of description length based on which the MDL-based procedure learns a BBN. Secondly, we point out that the difference between the MDL-based and Cooper and Herskovits procedures is essentially in the priors rather than in the approaches (MDL and Bayesian), and recommend a class of priors from which the formula is obtained. Finally, we show as a merit of using the formula that a modified version of the Chow and Liu algorithm is obtained. The modified algorithm finds a set of trees rather than a spanning tree based on the MDL principle.

In this paper, the computational issue in the problem of learning Bayesian belief networks (BBNs) based on the minimum description length (MDL) principle is addressed. Based on an asymptotic formula of description length, we apply the branch and bound technique to finding true network structures. The resulting algorithm searches considerably saves the computation yet successfully searches the network structure with the minimum value of the formula. Thus far, there has been no search algorithm that finds the optimal solution for examples of practical size and a set of network structures in the sense of the maximum posterior probability, and heuristic searches such as K2 and K3 trap in local optima due to the greedy nature even when the sample size is large. The proposed algorithm, since it minimizes the description length, eventually selects the true network structure as the sample size goes to infinity.

A algorithm for designing a pattern classifier, which uses MDL criterion and a binary data structure, is proposed. The algorithm gives a partitioning of the range of the multi-dimensional attribute and gives an estimated probability model for this partitioning. The volume of bins in this partitioning is upper bounded by ο((log N/N)K/(K+2)) almost surely, where N is the length of training sequence and K is the dimension of the attribute. The convergence rates of the code length and the divergence of the estimated model are asymptotically upper bounded by ο((log N/N)2/(K+2)). The classification error is asymptotically upper bounded by ο((log N/N)1/(K+2)). Simulation results for 1-dimensional and 2-dimensional attribute cases show that the algorithm is practically efficient.

This paper proposes a practical training algorithm for artificial neural networks, by which both the optimally pruned model and the optimally trained parameter for the minimum prediction error can be found simultaneously. In the proposed algorithm, the conventional information criterion is modified into a differentiable function of weight parameters, and then it is minimized while being controlled back to the conventional form. Since this method has several theoretical problems, its effectiveness is examined by computer simulations and by an application to practical ultrasonic image reconstruction.

This paper presents a method for spatial and temporal segmentation of long image sequences which include multiple independently moving objects, based on the Minimum Description Length (MDL) principle. By obtaining an optimal motion description, we extract spatiotemporal (ST) segments in the image sequence, each of which consists of edge segments with similar motions. First, we construct a family of 2D motion models, each of which is completely determined by its specified set of equations. Then, based on these sets of equations we formulate the motion description length in a long sequence. The motion state of one object at one moment is determined by finding the model with shortest description length. Temporal segmentation is carried out when the motion state is found to have changed. At the same time, the spatial segmentation is globally optimized in such a way that the motion description of the entire scene reaches a minimum.

This paper's main objective is to clearly describe the construction of a universal code for minimizing Davisson's minimax redundancy in a range where the true model and stochastic parameters are unknown. Minimax redundancy is defined as the maximum difference between the expected persymbol code length and the per-symbol source entropy in the source range. A universal coding scheme is here formulated in terms of the weight function, i.e., a method is presented for determining a weight function which minimizes the minimax redundancy even when the true model is unknown. It is subsequently shown that the minimax redundancy achieved through the presented coding method is upper-bounded by the minimax redundancy of Rissanen's semi-predictive coding method.

This paper proposes a model for learning non-parametric densities using finite-dimensional parametric densities by applying Yamanishi's stochastic analogue of Valiant's probably approximately correct learning model to density estimation. The goal of our learning model is to find, with high probability, a good parametric approximation of the non-parametric target density with sample size and computation time polynomial in parameters of interest. We use a learning algorithm based on the minimum description length (MDL) principle and derive a new general upper bound on the rate of convergence of the MDL estimator to a true non-parametric density. On the basis of this result, we demonstrate polynomial-sample-size learnability of classes of non-parametric densities (defined under some smoothness conditions) in terms of exponential families with polynomial bases, and we prove that under some appropriate conditions, the sample complexity of learning them is bounded as O((1/ε)(2r1)/2r1n(2r1)/2r(1/ε)(1/ε)1n(1/δ) for a smoothness parameter r (a positive integer), where ε and δ are respectively accuracy and confidence parameters. Futher, we demonstrate polynomial-time learnability of classes of non-parametric densities (defined under some smoothness conditions) in terms of histogram densities with equal-length cells, and we prove that under some appropriate condition, the sample complexity of learning them is bounded as O((1/ε)3/21n3/2(1/ε)(1/ε)1n(1/δ)).