Scalable architecture for word HMM-based speech recognition and VLSI implementation in complete system

This paper describes a scalable architecture for real-time speech recognizers based on word hidden Markov models (HMMs) that provide high recognition accuracy for word recognition tasks. However, the size of their recognition vocabulary is small because its extremely high computational costs cause long processing times. To achieve high-speed operations, we developed a VLSI system that has a scalable architecture. The architecture effectively uses parallel computations on the word HMM structure. It can reduce processing time and/or extend the word vocabulary. To explore the practicality of our architecture, we designed and evaluated a complete system recognizer, including speech analysis and noise robustness parts, on a 0.18-/spl mu/m CMOS standard cell library and field-programmable gate array. In the CMOS standard-cell implementation, the total processing time is 56.9 /spl mu/s/word at an operating frequency of 80 MHz in a single system. The recognizer gives a real-time response using an 800-word vocabulary.     

Synchronized transform-limited operation of 10-GHz colliding pulse mode-locked laser

We report a 10-GHz colliding pulse mode-locked laser fabricated with integrated active-passive waveguides. The laser fabrication adopted a deep reactive ion etching and single-step metal-organic vapor phase epitaxy regrowth process for forming the buried heterostructure waveguide. Clean output pulses resulted from laterally tilting the active-passive interface and effectively suppressing residual back-reflections at the interface. Hybrid mode-locking resulted in a synchronized transform-limited sech/sup 2/optical waveform. Pulsewidth, chirp, timing jitter, and frequency-locking range were investigated through systematic device biasing condition optimization.     

VBR Video Source Characterization and a Practical Hierarchical Model

There are many ways to build up traffic models for VBR video sources. A frequently applied methodology is to use mathematical analysis based on realistic assumptions to set up a source model that generates traffic according to a stochastic process. In this case, the critical issue is the validation of the synthetic trace by comparing statistics to results obtained from measurements on the real source. In this paper, we choose a different and more practical approach to model the behavior of the real traffic source. Our model building philosophy is that we analyze and understand what happens with the video information on its way from the ingress to the multimedia terminal to the egress of the network card. Throughout this journey the information is processed by several mechanisms and we build an empirical model step by step based on our measurement-based observations. Besides understanding the traffic generation procedure, statistical analysis of VBR traffic traces captured from a number of video sequences was also carried out in several scenarios. Using the knowledge of encoding, encapsulation and scheduling processes and results of the trace analysis, a hierarchical source model is set up for modeling the multimedia terminal. Thereby our model imitates the generation of video frames and the inner working of each level of protocol hierarchy and tries to reproduce the complex behavior of the real source. We use the leaky bucket analysis for verification of the model in order to capture directly the behavior of the traffic in a queue.     

Budding of fowlpox and pigeonpox viruses at the surface of infected cells

Chick embryo fibroblasts and chorioallantoic membranes infected with fowlpox virus (FWPV) or pigeonpox virus (PPV) were examined by transmission and scanning electron microscopy. Immature virus particles were observed in finely granular areas, i.e. virus factories, of the cytoplasm. These particles had various forms depending on their stages of development. Many tubular structures were also seen in these regions. Mature virus particles with ellipsoidal or brick-shaped forms enclosing electron-dense cores were detected throughout the cytoplasm. Notably, there was a high frequency of virus budding at the cell surface, but only occasional virus wrapping in the cytoplasm. Another remarkable feature of the infected cells was accumulation of many virions just beneath the plasma membrane, indicating that this phenomenon is closely related to virus budding. Based on the observed frequency of budding, this mechanism seems to be the predominant way for FWPV and PPV to exit the cell.     

Room temperature continuous-wave operation of GaInNAs/GaAs VCSELsgrown by chemical beam epitaxy with output power exceeding 1 mW

The authors present a near 1200 nm wavelength vertical-cavity surface-emitting laser (VCSEL) the active region of which is formed by GaInNAs/GaAs quantum wells grown by chemical beam epitaxy. Room temperature continuous-wave operation has been demonstrated with a high slope efficiency of 0.23 W/A, a high output power of over 1.0 mW, and a singlemode output power of 0.34 mW. The fabricated devices with different aperture sizes show low threshold currents of 1.2-2.1 mA and a record low threshold current density of 2.6 kA/cm² in GaInNAs VCSELs     

On the frame-error rate of concatenated turbo codes

Turbo codes with long frame lengths are usually constructed using a randomly chosen interleaver. Statistically, this guarantees excellent bit-error rate (BER) performance but also generates a certain number of low weight codewords, resulting in the appearance of an error floor in the BER curve. Several methods, including using an outer code, have been proposed to improve the error floor region of the BER curve. We study the effect of an outer BCH code on the frame-error rate (FER) of turbo codes. We show that additional coding gain is possible not only in the error floor region but also in the waterfall region. Also, the outer code improves the iterative APP decoder by providing a stopping criterion and alleviating convergence problems. With this method, we obtain codes whose performance is within 0.6 dB of the sphere packing bound at an FER of 10^-6     

International direct dialing quality in a competitive transitionaltelecommunications market

The introduction of resale-based competition in international direct dialing services in January 1999 triggered a round of extremely fierce competition in Hong Kong's IDD market. In response, both the incumbent operator and new entrants had to adopt aggressive strategies to defend or gain market share. This article reports on an intensive experiment on IDD quality provided by the major IDD operators in Hong Kong after this phase of deregulation. Based on 1790 successful IDD calls to the 10 most popular destinations from Hong Kong, the IDD quality of the major operators was benchmarked. The experiment revealed some interesting findings with significant implications for telecommunications deregulation     

Dictionary design for text image compression with JBIG2

The JBIG2 standard for lossy and lossless bilevel image coding is a very flexible encoding strategy based on pattern matching techniques. This paper addresses the problem of compressing text images with JBIG2. For text image compression, JBIG2 allows two encoding strategies: SPM and PM&S. We compare in detail the lossless and lossy coding performance using the SPM-based and PM&S-based JBIG2, including their coding efficiency, reconstructed image quality and system complexity. For the SPM-based JBIG2, we discuss the bit rate tradeoff associated with symbol dictionary design. We propose two symbol dictionary design techniques: the class-based and tree-based techniques. Experiments show that the SPM-based JBIG2 is a more efficient lossless system, leading to 8% higher compression ratios on average. It also provides better control over the reconstructed image quality in lossy compression. However, SPM's advantages come at the price of higher encoder complexity. The proposed class-based and tree-based symbol dictionary designs outperform simpler dictionary formation techniques by 8% for lossless and 16-18% for lossy compression     

Identification in the presence of side information with applicationto watermarking

Watermarking codes are analyzed from an information-theoretic viewpoint as identification codes with side information that is available at the transmitter only or at both ends. While the information hider embeds a secret message (watermark) in a covertext message (typically, text, image, sound, or video stream) within a certain distortion level, the attacker, modeled here as a memoryless channel, processes the resulting watermarked message (within limited additional distortion) in attempt to invalidate the watermark. In most applications of watermarking codes, the decoder need not carry out full decoding, as in ordinary coded communication systems, but only to test whether a watermark at all exists and if so, whether it matches a particular hypothesized pattern. This fact motivates us to view the watermarking problem as an identification problem, where the original covertext source serves as side information. In most applications, this side information is available to the encoder only, but sometimes it can be available to the decoder as well. For the case where the side information is available at both encoder and decoder, we derive a formula for the identification capacity and also provide a characterization of achievable error exponents. For the case where side information is available at the encoder only, we derive upper and lower bounds on the identification capacity. All characterizations are obtained as single-letter expressions     

Monolithically integrated 780-nm-band high-power and 650-nm-bandlaser diodes with real refractive index guided self-aligned structure

780-nm-band high-power and 650-nm-band laser diodes (LDs) with real refractive index guided self-aligned (RISA) structures are monolithically integrated for the first time. High-power and fundamental transverse mode operation at an output power of 100-mW continuous wave (CW) up to 80°C is attained for the 780-nm-band LD. For the 650-nm-band LD, high temperature and fundamental transverse mode operation at an output power of 10-mW CW up to 80°C is obtained