In order to inspect the valid speech interface, we have developed an emergency voice call-response system whereby when a message is spoken into a telephone in the event of an accident or a disaster, the employees wanted are automatically reached and called on to respond to the call. A word recognition algorithm in this system is using a non-linear matching processing technique from the reason that need to carry out same time parallel processing of multiple-line, and recognition object word number is small-vocabulary (32 words). And we have developed a new method whereby computer recognizable standard patterns are generated by interval estimation processing limited small voice data. We were able to identify that validity of speech interface by field testing result.
This paper describes about the description of this new method, and validity of speech interface for electric power enterprise.

抄録全体を表示

We can communicate with a visually impaired person through spoken language. However, it is extremely difficult for an aurally or verbally impaired person who understands by sign language, finger characters, or similar means to communicate with someone who does not understand such means. Conversion with writing is highly effective for communication, but it has a problem with speed.
With this background, the authors manufactured a communication support system that uses aa computer to pronounce submitted finger characters in real-time mode with a data glove and a magnetic sensor to measure the position and orientation.
In addition, fuzzy reasoning was introduced to cope with variations in the measured value using the fuzziness of the membership function. The recognition rate at the cluster level indicated a value of 90% or higher. As a result of using this determined hand form to recognize a finger character based on Enter angle data, a recognition rate of 90% or higher could always be obtained for the same subject, indicating valid application of fuzzy reasoning.

抄録全体を表示

We developed a globe information system for the blind using a color TV camera. The presentation of data on a tactile geographical globe, for blind people, is necessarily difficult, due to the limited space for Braille characters. To overcome this difficulty a TV camera was used to detect the position of the user's finger, with a speech synthesizer verbalising the information relating to the indicated point. Finger positioning was detected within an error of 5 degrees in latitude and 10 degrees in longitude. Total processing time were 4 sec. and 7 sec. using a work station (SS 2) and a personal computer(486, 50MHz) respectively. The voice information corresponding to the selected point was confirmed satisfactorily. Note that, as no sensors are required on the globe, then no modifications to the globe are needed. Although the cost of the color TV camera, image processing board and personal computer is high, the system has proved suitable for gathering information from a three dimensional object.

抄録全体を表示

There has been an increase in interest in the development of human-interface system that incorporates multimodal input-methods.
In this study, an Automaton-Controlled One Pass Viterbi program was developed to recognize speaker independent continuous speech. The base-line recognition mechanism is based on the One Pass Viterbi algorithm using the concatenation of syllable HMMs(Hidden Markov Models). In this program, if the dictionary and automaton information files based on the definition of finite state automaton are prepared, all kinds of sentences for the automaton can be accepted.
As an application of the speech recognition, the input methods by speech and by touch-screen were studied. Since these two methods are complementary each other, they can be used together to create more userfriendly human interface. To illustrate this, a multi-modal robot control simulation system was built on the Sun Sparc10 (using the attached AD converter with 11.025kHz sampling rate). To evaluate this system, a robot control simulation system using only the speech input method and a simulation system using only the touch-screen input method were also developed. After evalvating these systems, it was found out that the multi-modal system incorporating both speech input and touch-screen input was able to cover up the deficiencies that were in the use of only touch-screen input and to improve speech recognition speed and error rate. Also, by incorporating both speech input and touch-screen input, the operational errors became infrequent and more kinds of useful operations could be performed as compared to using only speech or only touch-screen inputs.

抄録全体を表示

With the proliferation of on-board information in recent years, automobile display devices, such as navigation system, have multiplied and instrument panels have become very complex. Still, a driver must be able to see and understand the information. For safe and comfortable driving, there are increasing demands for improving visibility of the navigation system. There have been various reports for visibility of the navigation system. But, most of the reports have been on the evaluation by human sense, and there has been few for the visibility evaluation in driving.
This paper describes the method for detecting the driver's eye movements with a single TV camera on dashboard, measuring the gaze duration and evaluating visibility in driving. In addition, the usefulness of the navigation system with voice route guidance compared with a conventional non-voice system has been demonstrated using the measuring system based on this method.

抄録全体を表示

Eye movements are realized as an electromechanism by controlling its optic axis according to the movement types of an object using artificial neural network. In order to realize basic optic axis movements, by which a moving object can be caught in a central pit of retina, an oculomotor machine is developed for independent horizontal movements of a head and an eyeball. The image output from the CCD camera used as a function of eyeball is fed to the control of a servomotor instead of muscles of eyeball to realize eye movements. The mechanism of head rotation is linked through bearings to the neck of an oculomotor machine. An accelerometer is used for a function of semicircular ducts in order to predict possible displacement of a head rotation. Theoculomotor machine is controlled for the realization of saccadic and smooth pursuit eye movements as autokinesis and compensatory eyemovement as a reflex with improvement of their control performances and automatic selection of an appropriate control law according to the movement types of an object, applying structuralized neural network information processing system

抄録全体を表示

The oscillation wavelength shift of a semiconductor laser in a magnetic field has been known since the early 60's. However, this research has only been made at very low temperatures and in a very strong magnetic field. The observed wavelength shift was toward a short wavelength side, which was well accounted for in terms of the Landau level. The wavelength shift in a relatively weak magnetic field at room temperature by using recently developed Fabry-Perot type laser diodes has been reported recently. The observed wavelength shift was toward a long wavelength side in certain types of laser diodes in a certain magnetic field direction. The shift mechanism, however, has not been confirmed yet. Therefore, the wavelength shift and the output laser power change in a magnetic field are measured precisely in this work. The longer wavelength side shift and the output laser power decrease occur simultaneously in a certain magnetic field direction. Since these changes observed in this work correspond to the changes observed in increasing temperature, it is assumed that the magnetic field increases temperature around the active layer of the semiconductor laser.

抄録全体を表示

strain-induced band offsets of the conduction band and valence subbands in Cd_xZn_1-xs/Zns superlattices have theoretically been calculated on the basis of the model-solid theory taking into account of spin-orbit energy (Δ₈₀). The conduction and valence band offsets have been estimated as a function of the well layer thickness and Cd composition (x), and as a result the valence band offsets which are derived from heavy-hole and light-hole bands increase with both decreasing the well layer thickness and Zn composition. Effect of external uniaxial compressive stress (ε) on the band offsets indicates that the energy difference between the heavy-hole band in CdZnS well and the light-hole band in zns barrier increases monotonously with applied stress. On the other hand, that of light-hole bands shows a quadratic energy shift as a function of uniaxial stress and has a minimum energy value around εΔ/₆₀. It is pointed out that a large energy difference in uniaxial deformation potential constants between CdS and ZnS shows a unique behaviour which significantly modifies the electronic states in the CdZns-based quantum well structures under uniaxial stress.

抄録全体を表示

The rule selection planning expert system, which selects a suitable dispatching rule from dispatching rules in knowledge-base according to planning process, is an influential approach to planning by computer. However, since knowledge of rule selection depends oil the dispatching rules, it is not easy to exchange the dispatching rules. This paper presents a new method for selection of the dispatching rules, named a status selection method. In this method, the most promising status is selected from tentative statuses that are generated by applying all dispatching rules. Since the dispatching rules and the knowledge of status selection are independent, it is easy to exchange the dispatching rules. From the results of the application of the proposed method to a simple flow shop problem and a job shop problem, we confirm thatt the status selection method is so effective.

抄録全体を表示