This paper describes continuous speech recognition incorporating the additional complement information, e.g., voice characteristics, speaking styles, linguistic information and noise environment, into HMM-based acoustic modeling. In speech recognition systems, context-dependent HMMs, i.e., triphone, and the tree-based context clustering have commonly been used. Several attempts to utilize not only phonetic contexts, but additional complement information based on context (factor) dependent HMMs have been made in recent years. However, when the additional factors for testing data are unobserved, methods for obtaining factor labels is required before decoding. In this paper, we propose a model integration technique based on general factor dependent HMMs for decoding. The integrated HMMs can be used by a conventional decoder as standard triphone HMMs with Gaussian mixture densities. Moreover, by using the results of context clustering, the proposed method can determine an optimal number of mixture components for each state dependently of the degree of influence from additional factors. Phoneme recognition experiments using voice characteristic labels show significant improvements with a small number of model parameters, and a 19.3% error reduction was obtained in noise environment experiments.

In this paper, we report the fabrication and device characteristics of InP/InGaAs double heterojunction bipolar transistors (DHBTs) with buried SiO2 wires. The SiO2 wires were buried in the collector and subcollector layers by metalorganic chemical vapor deposition toward reduction of the base-collector capacitance under the base electrode. A current gain of 22 was obtained at an emitter current density of 1.25 MA/cm2 for a DHBT with an emitter width of 400 nm. The DC characteristics of DHBTs with buried SiO2 wires were the same as those of DHBTs without buried SiO2 wires on the same substrate. A current gain cutoff frequency (fT) of 213 GHz and a maximum oscillation frequency (fmax) of 100 GHz were obtained at an emitter current density of 725 kA/cm2.

We fabricate a 763-nm laser module based on second-harmonic generation using a direct-bonded quasi-phase-matched LiNbO3 ridge waveguide. We obtained a 0.84-mW output of 763 nm light using a 1526-nm distributed-feedback laser diode. We also demonstrate O2 gas detection using the module output.

We propose and demonstrate a novel ladder interferometric filter that exhibits flat-topped spectral response for use in wavelength-division-multiplexing (WDM) based photonic networks. We numerically analyze the flattened spectral response in a ladder-type filter by modifying the transfer matrix of ladder interferometer. Conventional parabolic-shaped and flat-topped-designed ladder interferometric filters are fabricated, and characterized. We demonstrate a flat-topped filter response in the fabricated device. The shape factor, which is defined by the ratio of -1 dB bandwidth to -10 dB bandwidth, is improved from 0.32 to 0.54. The tunability and the increase in filter extinction ratio of the proposed device are also discussed.

We investigated collector current spreading in InGaAs single heterojunction bipolar transistors (SHBTs) having a collector thickness of 75 nm. SHBTs were fabricated with three different emitter widths -- 200, 400, and 600 nm -- and the highest cutoff frequency that was obtained was 468 GHz. The relationship between the current density at the highest cutoff frequency and the emitter width could not be used to estimate the current spreading because it was independent of the collector-base voltage. However, the relationship between the current density with the increase in the total collector-base capacitance and the emitter width indicates current spreading in the collector. The current spreading was estimated to be approximately 90 nm.

This paper describes recent progress in research on wavelength converters that employ quasi-phase-matched LiNbO3 (QPM-LN) waveguides. The basic structure and operating principle of these devices are presented. The conversion efficiency in difference frequency generation (DFG), second harmonic generation (SHG) and an SHG/DFG cascade scheme are explained. Device fabrication technologies such as periodic poling, and those used for annealed proton-exchanged (APE) waveguides, and direct bonded waveguides are introduced. An APE waveguide is used to demonstrate the wavelength conversion of broadband (> 1 Tbit/s) WDM signals. The low penalty conversion of high-speed (40 Gbit/s) based WDM signals is also reported. Excellent resistance to photorefractive damage in a direct bonded waveguide is presented. This high level of resistance enabled highly efficient wavelength conversion. A new design concept is introduced for a multiple QPM device based on the continuous phase modulation of a periodically poled structure. This multiple QPM device enables the variable wavelength conversion of WDM signals. High-speed wavelength switching between ITU-T grid wavelengths using a finely tuned multiple QPM device is also reported. QPM-LN based wavelength converters have several advantages, including the ability to convert high-speed signals of 1 THz or greater, no signal-to-noise (S/N) ratio degradation, no modulation format dependence, and they are capable of the simultaneous conversion of broadband WDM channels. They will therefore be key devices in future photonic networks.

Apodised chirped gratings based on InGaAsP/InP deep-ridge waveguides with vertical-groove surface gratings were fabricated. Reflectivity ripple and group delay ripple were reduced from around 4 dB to 1 dB and from around 5 ps to 2 ps, respectively, by apodisation over a wavelength range of around 20 nm.

Asynchronous optical packet switching (OPS) is a promising solution to support the continuous growth of transmission capacity demand. It has been, however, quite difficult to implement key functions needed at the node of such networks with all-optical approaches. We have proposed a new optoelectronic system composed of a packet-by-packet optical clock-pulse generator (OCG), an all-optical serial-to-parallel converter (SPC), a photonic parallel-to-serial converter (PSC), and CMOS circuitry. The system makes it possible to carry out various required functions such as buffering (random access memory), optical packet compression/decompression, and optical label swapping for high-speed asynchronous optical packets.

In developing the SXO operating system for the SURE SYSTEM 2000 continuous operation system, we aimed to create an unprecedentedly high software and hardware fault tolerance. We devised a fault tolerant architecture and various methodologies to ensure fault tolerance. We implemented these techniques systematically throughout operating system development. In the design stage, we developed a design methodology called the recovery process chart to verify that recovery mechanisms were complete. In the manufacturing stage, we applied the concept of critical routes to recovery and other processes essential to high dependability. We also developed a method of finding critical routes in a recovery process chart. In the test stage, we added an artificial software fault injection mechanism to the operating system. It generates various reproducible errors at appropriate times and reduces the number of personnel needed for test, making system reliability evaluation easy.

In recent years, with the improvement of health awareness, people have paid more and more attention to proper meal. Existing research has shown that a proper meal can help people prevent lifestyle diseases such as diabetes. In this research, by attaching sensors to the tableware, the information during the meal can be captured, and after processing and analyzing it, the meal information, such as time and sequence of meal, can be obtained. This paper introduces how to use supervised learning and multi-instance learning to deal with meal information and a detailed comparison is made. Three supervised learning algorithms and two multi-instance learning algorithms are used in the experiment. The experimental results showed that although the supervised learning algorithms have achieved good results in F-score, the multi-instance learning algorithms have achieved better results not only in accuracy but also in F-score.

We propose a burst optical packet generator based on a novel photonic parallel-to-serial conversion scheme, and demonstrate 40-Gbit/s 16-bit optical packet generation from 16-ch parallel low-voltage TTL data streams. It consists of electrical 4:1 parallel-to-serial converters that employ InP metal-semiconductor-metal photodetectors, and an optical time-domain multiplexer with electroabsorption modulators. The proposed optical packet generator is suitable for burst optical packet generation and overcomes the electronic bandwidth limitation, which is prerequisite for achieving high-speed photonic packet switched networks. In addition, it can be driven by simple low-cost low-power CMOS logic circuits, and is compact and extensible in terms of the number of input channels due to the effective combination of electrical and optical multiplexing.