In this paper, we address issues in improving hands-free speech recognition performance in different car environments using multiple spatially distributed microphones. In the previous work, we proposed the multiple linear regression of the log spectra (MRLS) for estimating the log spectra of speech at a close-talking microphone. In this paper, the concept is extended to nonlinear regressions. Regressions in the cepstrum domain are also investigated. An effective algorithm is developed to adapt the regression weights automatically to different noise environments. Compared to the nearest distant microphone and adaptive beamformer (Generalized Sidelobe Canceller), the proposed adaptive nonlinear regression approach shows an advantage in the average relative word error rate (WER) reductions of 58.5% and 10.3%, respectively, for isolated word recognition under 15 real car environments.

Machine transliteration is an automatic method to generate characters or words in one alphabetical system for the corresponding characters in another alphabetical system. Machine transliteration can play an important role in natural language application such as information retrieval and machine translation, especially for handling proper nouns and technical terms. The previous works focus on either a grapheme-based or phoneme-based method. However, transliteration is an orthographical and phonetic converting process. Therefore, both grapheme and phoneme information should be considered in machine transliteration. In this paper, we propose a grapheme and phoneme-based transliteration model and compare it with previous grapheme-based and phoneme-based models using several machine learning techniques. Our method shows about 1378% performance improvement.

A fully integrated broadband distributed frequency tripler, periodically loaded with HBV devices, has been designed and fabricated and has demonstrated the generation of a broad range of output frequencies of up to 570 GHz. Key to the design is the principle that the entire frequency tripler circuit is produced monolithically and incorporates novel HBV devices electrically and mechanically interconnected by a thin low-loss SU-8 membrane. With the device fabrication approach used, the novel HBV devices are able to produce a higher capacitance-voltage swing ratio whilst simultaneously minimizing the device series and contact resistances to achieve the optimum conversion efficiency. The entire concept of this work was to design a cost effective fully integrated waveguide package, with the frequency tripler circuit mounted at the E-plane of a micromachined waveguide which was constructed with stepped height and width to prevent the propagation of higher order modes inside the waveguide sections. The micromachined waveguide sections exhibit high dimensional accuracy and a good surface finish which is necessary for the efficient propagation of high frequency signals. The frequency tripler circuit and the accompanying micromachined waveguide sections are mounted in a specifically designed metal test fixture to form a compact and cost-effective subcomponent with great commercial potential for broadband harmonic generation of up to terahertz frequencies. This paper presents the design methodology and techniques used to produce the frequency tripler package, together with some initial measurement results.

In order to boost the translation quality of corpus-based MT systems for speech translation, the technique of splitting an input utterance appears promising. In previous research, many methods used word-sequence characteristics like N-gram clues among splitting positions. In this paper, to supplement splitting methods based on word-sequence characteristics, we introduce another clue using similarity based on edit-distance. In our splitting method, we generate candidates for utterance splitting based on N-grams, and select the best one by measuring the utterance similarity against a corpus. This selection is founded on the assumption that a corpus-based MT system can correctly translate an utterance that is similar to an utterance in its training corpus. We conducted experiments using three MT systems: two EBMT systems, one of which uses a phrase as a translation unit and the other of which uses an utterance, and an SMT system. The translation results under various conditions were evaluated by objective measures and a subjective measure. The experimental results demonstrate that the proposed method is valuable for the three systems. Using utterance similarity can improve the translation quality.

Effective trading with given pattern-based multi-predictors of stock price needs an intelligent asset allocation strategy. In this paper, we study a method of dynamic asset allocation, called the meta policy, which decides how much the proportion of asset should be allocated to each recommendation for trade. The meta policy makes a decision considering both the recommending information of multi-predictors and the current ratio of stock funds over the total asset. We adopt evolutionary computation to optimize the meta policy. The experimental results on the Korean stock market show that the trading system with the proposed meta policy outperforms other systems with fixed asset allocation methods.

The efficiency of algorithms managing data caches has a major impact on the performance of systems that utilize client-side data caching. In these systems, two versions of data can be maintained without additional overhead by exploiting the replication of data in the server's buffer and clients' caches. In this paper, we present a new cache consistency algorithm employing versions: Two Versions-Callback Locking (2V-CBL). Our experimental results indicate that 2V-CBL provides good performance, and in particular outperforms a leading cache consistency algorithm, Asynchronous Avoidance-based Cache Consistency, when some clients run only read-only transactions.

Simultaneous recordings of eight channel surface myoelectric signals (EMGs) of the biceps brachii muscles of seven subjects were measured in isovelocity elbow flexion against constant load torque. The velocity was 10, 15, 20 and 25 degree/s and the load torque was 5-15 % of the torque obtained at the maximum voluntary contraction (MVC). Individual motor units were identified from the eight-channel surface EMG, by tracking the waveform change which originated from the change of relative position of muscle fiber and electrode. In the low-load (5 and 7% MVC) experiment, 36 examples of recruitment and 22 examples of derecruitment were measured. In the middle-load (10 and 15% MVC) experiment, most of the motor units did not show an obvious change in the firing rate with the elbow joint angle. Average of the firing rates of all the motor units measured at the elbow angle of 0 to 120 degree (13.3-14.7 Hz) did not depend on flexion velocity between 10 to 25 degree/s. It was concluded that the firing rates of the activated MUs were almost constant and that some MUs were recruited and derecruited during the isovelocity flexion movements. These are the first findings.

This paper discusses several practical issues related to the provision of video-on-demand (VOD) services, focusing on retrieval of video data from disk on the server. First, with regard to system design, the pros and cons of cycle-based scheduling algorithms for VOD servers are compared, and an adequate policy according to system configuration is presented. Second, we present a way to tune the cycle-based scheduling algorithm so that it maximizes profit. Third, a method to overcome the cons of cycle-based scheduling algorithms is proposed, and its cost is analyzed.

We have developed a traveling-wave optical modulator using an n-i-n heterostructure fabricated on an InP substrate. The modulation characteristics are studied theoretically and experimentally. We obtained an extremely small π voltage (Vπ) of 2.2 V, even for a short signal-electrode length of 3 mm. We confirmed a wide frequency bandwidth and clearly open eye diagrams at 40 Gbit/s.

We investigate the error exponent in the lossy source coding with a fidelity criterion. Marton (1974) established a formula of the reliability function for the stationary memoryless source with finite alphabet. In this paper, we consider a stationary memoryless source assuming that the alphabet space is a metric space and not necessarily finite nor discrete. Our aim is to prove that Marton's formula for the reliability function remains true even if the alphabet is general.

Peer-to-peer (P2P) web caching has been studied recently as it can exploit local caches of peers for web caching without additional infrastructure. However, dynamic join/leave behaviors or local caching strategies of the peers due to their autonomy in a P2P network may limit the performance of P2P web caching. To overcome these limitations, we propose an effective directory-based P2P web caching system under dynamic participation of peers. We introduce the object lifetime in a P2P network considering the lifetimes of both an object in the local cache of a peer and a peer who owns the object, and utilize this object lifetime information for neighbor selection and storage management in the P2P web caching framework. For the neighbor selection, the proposed system utilizes the object lifetimes in selecting accurately a neighbor who would still retain the requested object and still remain in the P2P network. To improve the storage management, the proposed system uses efficiently the entire cache storage of the P2P network in such a way that the object is stored selectively in the local cache of the peer who requested it, considering the object lifetime. The trace-driven simulation results show that the proposed system has higher accuracy and fewer redirection failures than the conventional directory-based P2P web caching system in the feasible P2P network.

The performance of en-route web caching mainly depends on where the caches are located and how the cache contents are managed. In this paper, we address the problem of proxy placement in en-route web caching for tree networks, i.e., computing the optimal locations for placing k web proxies in a network such that some specified objectives are achieved. Based on our proposed model, we formulate this problem as an optimization problem and compute the optimal locations using a computationally efficient dynamic programming-based algorithm. We also extend our solution for tree networks to solve the same problem for autonomous systems. Finally, we implement our algorithms and evaluate our model on several performance metrics through extensive simulation experiments. We also compare the performance of our model with the best available heuristic KMPC model, as well as the random proxy placement model. The implementation results show that our model outperforms all the other models with respect to all performance metrics considered. The average improvements of our model over the KMPC model and the random proxy placement model are about 31.9 percent and 58.6 percent in terms of all the performance metrics considered.

As the transistor feature sizes and threshold voltages reduce, leakage energy consumption has become an inevitable issue for high-performance microprocessor designs. Since on-chip caches are major contributors of the leakage, a number of researchers have proposed efficient leakage reduction techniques. However, it is still not clear that 1) what kind of algorithm can be considered and 2) how much they have impact on energy and performance. To answer these questions, we explore run-time cache management algorithm, and evaluate the energy-performance efficiency for several alternatives.

This paper deals with a TE plane wave reflection and transmission from a one-dimensional random slab by means of the stochastic functional approach. The relative permittivity of the random slab is written by a Gaussian random field in the vertical direction with finite thickness, and is uniform in the horizontal direction with infinite extent. An explicit form of the random wavefield is obtained in terms of a Wiener-Hermite expansion with approximate expansion coefficients (Wiener kernels) under a small fluctuation case. By using the first three terms of the random wavefield representation, the optical theorem is illustrated in figures for several physical parameters. It is then found that the optical theorem holds with good accuracy.

A novel cache-based packet-processing-engine (PPE) architecture that achieves low-power consumption and high packet-processing throughput by exploiting the nature of network traffic is proposed. This architecture consists of a processing-unit array and a bit-stream manipulation path called a burst stream path (BSP) that has a special cache mechanism called a process-learning cache (PLC). Network packets, which have the same information in their header, appear repeatedly over a short time. By exploiting that nature, the PLC memorizes the packet-processing method with all results (i. e. , table lookups), and applies it to other packets. The PLC enables most packets to skip the execution at the processing-unit array, which consumes high power. As a practical implementation of the cache-based PPE architecture, P-Gear was designed. In particular, P-Gear was compared with a conventional PPE in terms of silicon die size and power consumption. According to this comparison, in the case of current 0.13-µm CMOS process technology, P-Gear can achieve 100-Gbps (gigabit per second) packet-processing throughput with only 36.5% of the die size and 32.8% of the power consumption required by the conventional PPE. Configurations of both architectures for the 1- to 100-Gbps throughput range were also analyzed. In the throughput range of 10-Gbps or more, P-Gear can achieve the target throughput in a smaller die size than the conventional PPE. And for the whole throughput range, P-Gear can achieve a target throughput at lower power than the conventional PPE.

We propose a new method both for automatically creating non-uniform, context-dependent HMM topologies, and selecting the number of mixture components based on the Variational Bayesian (VB) approach. Although the Maximum Likelihood (ML) criterion is generally used to create HMM topologies, it has an over-fitting problem. Recently, to avoid this problem, the VB approach has been applied to create acoustic models for speech recognition. We introduce the VB approach to the Successive State Splitting (SSS) algorithm, which can create both contextual and temporal variations for HMMs. Experimental results indicate that the proposed method can automatically create a more efficient model than the original method. We evaluated a method to increase the number of mixture components by using the VB approach and considering temporal structures. The VB approach obtained almost the same performance as the smaller number of mixture components in comparison with that obtained by using ML-based methods.

This paper presents speech-driven Web retrieval models which accept spoken search topics (queries) in the NTCIR-3 Web retrieval task. The major focus of this paper is on improving speech recognition accuracy of spoken queries and then improving retrieval accuracy in speech-driven Web retrieval. We experimentally evaluated the techniques of combining outputs of multiple LVCSR models in recognition of spoken queries. As model combination techniques, we compared the SVM learning technique with conventional voting schemes such as ROVER. In addition, for investigating the effects on the retrieval performance in vocabulary size of the language model, we prepared two kinds of language models: the one's vocabulary size was 20,000, the other's one was 60,000. Then, we evaluated the differences in the recognition rates of the spoken queries and the retrieval performance. We showed that the techniques of multiple LVCSR model combination could achieve improvement both in speech recognition and retrieval accuracies in speech-driven text retrieval. Comparing with the retrieval accuracies when an LM with a 20,000/60,000 vocabulary size is used in an LVCSR system, we found that the larger the vocabulary size is, the better the retrieval accuracy is.

An efficient, scalable speech recognition architecture combining topic detection and topic-dependent language modeling is proposed for multi-domain spoken language systems. In the proposed approach, the inferred topic is automatically detected from the user's utterance, and speech recognition is then performed by applying an appropriate topic-dependent language model. This approach enables users to freely switch between domains while maintaining high recognition accuracy. As topic detection is performed on a single utterance, detection errors may occur and propagate through the system. To improve robustness, a hierarchical back-off mechanism is introduced where detailed topic models are applied when topic detection is confident and wider models that cover multiple topics are applied in cases of uncertainty. The performance of the proposed architecture is evaluated when combined with two topic detection methods: unigram likelihood and SVMs (Support Vector Machines). On the ATR Basic Travel Expression Corpus, both methods provide a significant reduction in WER (9.7% and 10.3%, respectively) compared to a single language model system. Furthermore, recognition accuracy is comparable to performing decoding with all topic-dependent models in parallel, while the required computational cost is much reduced.

This paper presents a design procedure of a directional coupler consisting of a twofold symmetric four-port circuit with four identical matching networks at each port. The intrinsic power-split ratio and the equivalent admittance of the directional coupler are formularized in terms of the eigenadmittances of the original four-port without the matching networks. These formulas are useful for judgment on the realizability of a directional coupler in a given circuit structure and for design of the matching networks. Actually, the present procedure is applied to designing various quadrature hybrids and directional couplers, and its practical usefulness as well as several new circuit structures are demonstrated.

The volume of mass unsolicited electronic mail, often known as spam, has recently increased enormously and has become a serious threat not only to the Internet but also to society. This paper proposes a new spam detection method which uses document space density information. Although the proposed method requires extensive e-mail traffic to acquire the necessary information, it can achieve perfect detection (i.e., both recall and precision is 100%) under practical conditions. A direct-mapped cache method contributes to the handling of over 13,000 e-mail messages per second. Experimental results, which were conducted using over 50 million actual e-mail messages, are also reported in this paper.