Abstracts of Magazine FUJITSU 1998-1 (VOL.49, NO.1)

Special Issue 1 : GRANPOWER Series

• High-Performance Technology for GRANPOWER 5000 Series

System performance of PC server depends on processors, chip sets and IO controllers.
To improve system performance and price-performance, best processors are implemented into GRANPOWER 5000 series.
The high performance technology originally developed for the mainframes is implemented into the GRANPOWER 5000 model 570. The GRANPOWER 5000 model 570 has achieved the performance by 20% faster than those of conventional products.

• High-Reliability Technology of Electro-Mechanical Design for GRANPOWER 5000 Series

PC servers are evolving from office servers into database servers of large-scale central systems. Because of these changes and the rapid growth of the PC server market, PC servers used in a normal office environment are now required to have broader functions and improved flexibility and high reliability. One of the requirements is minimal downtime. Other important one is high-reliability cabinets that allow easy maintenance and have advanced safety features.
Therefore, Fujitsu has made various improvements in the designs of its cabinets. Some examples of these improvements are power unit redundancy, hot pluggable HDDs, high-efficiency cooling fans, enhanced system maintainability, and reductions in physical size.

• Hardware of GRANPOWER 6000 Series

Network computing has become more popular in recent years, and the business servers that have been used as stand-alone systems are now being repositioned as mission-critical application servers for networks.
The GRANPOWER 6000 series is the newest line of Fujitsu business servers with the latest Pentium processors. The series is completely compatible with existing resources, including OSs, and makes it possible to cope with a rapidly expanding network using a uniform architecture. The new series is founded on the knowledge and techniques we have gained over the years and features various new functions.
This paper mainly describes the high reliability of the GRANPOWER 6000 series and the technologies that enable the series, with its new processors, to inherit existing resources.

• High-Performance Technology for GRANPOWER 6000 Series

The GRANPOWER 6000 series is a highly efficient and reliable business server with the latest PentiumPro processor. The OS of GRANPOWER 6000 series is fully compatible with the OS of K series. Application programs of K series completely can run on GRANPOWER 6000 series.
We developed the emulator that enabled the OS and application programs of K series to run on PentiumPro processor. We developed the dynamic OCT (Object Code Translation) technology at the same time. The dynamic OCT technology is an advanced emulation technology that translates the object code of K series to the object code of PentiumPro and achieves performance improvement.
This document mainly explains the dynamic OCT technology，which is the core technology of the GRANPOWER 6000 series.

• High-Performance Technology for GRANPOWER 7000 Series

This paper gives an overview of the GRANPOWER 7000 model 400 and model 600 UNIX server systems with 64-bit SPARC processors. They provide the scalable high performance and the high reliability for mission-critical applications. The high-end system has eight processors and 8-way memory sub-systems connected via a crossbar-switch.
Since each processor can access memory independently，these systems boast a high performance in proportion to the number of processors they contain; they have achieved one of the highest performance systems in the world. The reliability of these systems has been enhanced by adding ECC to the data path between the processor module and memory subsystem and to the DTAG, which is a control memory for checking the cache status.

• High-Reliability Technology for GRANPOWER 7000 Series

This paper gives an overview of the high reliability of the GRANPOWER 7000 series of UNIX business server systems designed for mission-critical transactions.
Although failover is not a new concept, it is an essential requirement for UNIX server systems and has therefore been incorporated into the GRANPOWER 7000 series in the form of Fujitsu's System Control Facility (SCF). The SCF is an independent supervisor of the host processor system and I/O subsystem. Its design concepts are fault-tolerance and flexible supervision. To achieve fault-tolerance, the series has spare fans and power supply units that can be hot-swapped with no server downtime.
Also, to further improve the supervisor facility, the internal/external temperature sensing and CPU monitoring have been enhanced.

• Software of GRANPOWER Series：SymfoWARE

Fujitsu has developed a hybrid database system called SymfoWARE Server which can manage any type of data. SymfoWARE Server provides a hybrid access feature, supporting relational type data and object type data.
The first version of SymfoWARE Server was released in October 1993. Currently, SymfoWARE Server is running on more than 5,000 systems, including UNIX servers (GRANPOWER 7000 series, Sfamily, etc.), business servers (GRANPOWER 6000 series, etc.) and PC servers (GRANPOWER 5000 series, etc.).
This paper describes how SymfoWARE Server fully meets many of the strict requirements of today's network information systems.

Special Issue 2 : Recent Technologies for Speech Synthesis and Speech Recognition

---

• Progress in Speech Synthesis and Speech Recognition Technologies

This paper outlines the basics of speech and recent technologies for speech synthesis and recognition to help readers understand the papers in this special issue on speech technologies. First, the basic terminology of speech science is explained. Then, the classification of technologies, outline of processing, current performance, and future problems concerning speech technologies are described.
Finally, we explain why the simple substitution of speech synthesis in place of a record-playback function or switches in place of speech recognition is not useful in actual applications and why we must focus on the inherent functions of speech communication using speech synthesis and speech recognition. This viewpoint regarding the use of speech technologies differs from the one held by mainstream speech technology engineers.

• High-Quality Speech Synthesis Technologies

This paper describes several high-quality speech synthesis technologies.
For a baseline text-to-speech system, we developed a word detection algorithm for accurate reading of Japanese text that includes kanji characters. This algorithm is based on a DP matching method and a 125,000-word dictionary. Also, we introduced a wave-packet concatenating method for articulated synthesized speech.
To read English words in Japanese text, we built a dictionary which assigns them Japanese pronunciations. For improved reading of Japanese text, we developed dictionaries that contain postal addresses in Japan and family and personal names of Japanese people.
Furthermore we propose a method of generating natural prosody (intonation and pauses) for sentences composed by inlaying variable words into slots of prepared sentence structures.

• Speech Synthesis Application Development Kit

The Speech Synthesis Application Development Kit enables users to add a speech synthesis facility to application programs. The kit has a speech synthesis library which performs three functions: language processing, wave processing, and dictionary control.
This paper gives an overview of these three functions.

• Development Tool for Telephony Information Service Using Speech Synthesis

The networking environment continues to evolve，and the telephone related segment of the Computer Telephony Integration (CTI) market is growing rapidly. The greatest expansion in the CTI market is occurring in the area of Interactive Voice Response (IVR) systems. These systems exploit the facts that the telephone is very easy to use and is an integral part of the IT infrastructure and therefore accessible to almost everybody most of the time (we are fast approaching the era of "one mobile phone or PHS per person").
The market is demanding new types of IVR applications and the means to develop them. In particular, advances in voice synthesis technology are being used to create more flexible and efficient IVR systems and applications.
To meet the market's demand for more sophisticated IVR systems and system development facilities, we have created the Voice Script voice and fax response system. This system allows the development of IVR applications incorporating both synthetic and natural speech messages. In the case of speech synthesis, new technologies have been employed to enable natural sounding voice synthesis from text data and file sources. This new technology facilitates rapid and high-productivity development of IVR systems in an open environment.

• Speech Synthesis Software for Home Use

Because of advances in personal computers and speech synthesis, many new software products include a speech synthesis function. Fujitsu has made a contribution in this area by developing a speech synthesis software package for home use called Oshaberi Mate.
This paper describes our work on speech synthesis software for home use. First, we describe the software's development background and the requirements for the home use market. Then, we describe some of the applications of speech synthesis. Finally, we look at some of the requests of users.

• Speech Synthesis System for Car Navigation

Due to the increase in the number of features provided by car electronics units, it is becoming important to develop Human Interface (HI) technologies that enable drivers to easily and safely use these features. To this end, we have developed a speech synthesis technology which enables a car driver to operate equipment without needing to look at a display. Then, using this technology, we developed a speech synthesis system for a car navigation system.
This paper outlines our speech synthesis system and describes several of its main features.

• Technologies for Continuous Speech Recognition

This paper describes some technologies for recognizing continuous speech such as whole sentences and word sequences.
Since continuous speech recognition is based on isolated word recognition, we first describe some technologies for isolated word recognition. Next, we describe some continuous speech recognition technologies and the methods used to reduce their computational complexity, which is the most significant problem in continuous speech recognition. Computational complexity can be reduced by converting sentence models into tree structures of word models and by using a best-first-search method to search the tree.
Then we describe the sentence recognition technology developed by Fujitsu called the Case-flag propagation method. This method is based on the kakari-uke grammar and is suitable for Japanese sentence recognition.
Lastly, we present some experimental results. We have achieved a 97.3 percent recognition rate for groups of 100 isolated words and a 98.0 percent sentence understanding rate for continuous speech recognition based on kakari-uke grammar.

• Technologies for Speech Recognition in Noisy Environment

Noise is a big obstacle to achieving speech recognition in a practical environment. This paper describes speech recognition technologies which extract text from noisy speech.
First, we explain how recognition systems without countermeasures for noise have a poor recognition performance for noisy speech. Next, we introduce some techniques for noisy speech recognition. Finally, we describe a speaker-independent word recognition system based on a spectrum subtraction technique and present some experimental results of speech recognition in a noisy environment.

• Speech Input Interface with Microphone Array

This paper describes a speech input interface designed for speech recognition in a noisy environment.
In a noisy environment, the interface detects and enhances the speech of a person speaking at a specified location. When the person stops speaking, the interface reduces its output level to zero. The interface uses a voice switch based on a sound source position detection algorithm and a directional microphone system consisting of a three-microphone array and a signal enhancement algorithm. The interface has performed well in various tests we have conducted.

• Multi-Modal Interface System Using Speech Recognition and Synthesis Technologies

This paper outlines the multi-modal interface system for personal computers that we have developed. It also describes the role of speech recognition and synthesis in the interface.
The multi-modal interface system consists of two key technologies we have developed. One is the general input commands, we call HyperWord, which does not depend on the type of input devices. The other is a language called Hyper Communication Language (HCL), which can define the environment for commanding a computer and translating user actions into HyperWord.
This system facilitates computer use. For example, if the user says "Read" and "Mail," the system immediately launches a mailer application so the user can read an e-mail. Thus, the user can operate a computer with two simple voice or gesture actions.

---

• Development of New OASYS

Since its debut in 1980, the OASYS word processor series has won praise for its easy and intuitive user interface and superior Japanese processing functions. However, the rapid growth of the personal computer market and platform standardization has stimulated the need to develop a new strategy for the OASYS series, which runs on a non-standard OS.
Another Fujitsu product, OASYS/Win, has advanced technologies such as WYSIWYG editing and OLE and has become an established Windows-based word processor. These two products have now been integrated into OASYS V5. This new product features an improved user interface, Internet/intranet support, and new advanced-Japanese processing functions.
This paper describes the new technologies and functions of OASYS V5.