A blind student writes and submits reports in Braille word processor, which is difficult for teachers to read. This study's purpose is to make a translator from Braille into mixed Kana-Kanji sentences for such teachers. Because Kanji has homonyms, it is not always possible to get correct results when converting. To overcome this difficulty, we used deep learning for translation. We built a training dataset composed from 15,000 pairs of Braille codes and mixed Kana-Kanji sentences, and a validation dataset. In training, we got an accuracy of 0.906 and a good Bleu score of 0.600. In validation, we found 5 mistaken words in selecting homonymous Kanji by examining translation mistakes from 100 pairs of the verification sentences. The choice of homonymous Kanji depends on the context. For decreasing such type of errors, it is necessary to introduce of translation of paragraphs by increasing the scale of the network model in deep learning, and to expand the network structure.

Recently, research on machine translation has been active, and research on the Transformer, which is a translation component of the open source library Tensor2Tensor, is also active. In July 2020, a development research institute called Open AI announced GPT − 3 (Generative Pretrained Transformer), a sentence generation language model, uses 175 billion parameters and is evaluated as an AI tool that generates sentences as if they were written by humans. Using this Transformer, we are also constructing a Braille translation system using a neural network. To translate Japanese into Braille, we devised a three-step process. The first is the step of converting a kana-kanji mixed sentence into a space-inserted sentence, the second is the step of converting the space-inserted sentence into a phonetic kana sentence, and the third is the step of converting the phonetic kana sentence into Braille. Earlier at ICMSEM 2019, we have presented the third step. This time, we have developed the first and second steps. That is, the input data are the kana − kanji mixed sentence of step 1, and the output is the phonetic kana sentence of step 2. The whole picture of this system and the steps developed this time is described.

Braille books follow processes of deteriorations different from ordinary printed books because visually impaired people read the sharp dot with the finger’s belly. In frequently read braille books, braille is dirty with aging, collapses, and holes open. Then, visually impaired people cannot read braille books. For braille books manually transcribed, the only method to republish them is to rebuild though the original is often and already disposed of. Now, we read degraded braille books with a scanner and restore them. In this paper, comparative experiments on what types of scanner are appropriate to successfully read degraded braille from old braille books are carried out. The candidates are two flatbed scanners, GT-S 650 and Plustek OpticBook 4800, and a stand scanner, ScanSnap SV600. Each scanner scans three types of braille papers, fresh books, used books and old books. Next, to recognize and identify dots in braille papers, we use two methods, the Haar classifier in OpenCV as image processing function and the deep learning with Google’s TensorFlow. For aiming at improving the recognition rate, we also use scanner options suitable for reading. Experimental results show that the flatbed scanner is suitable for reading.

We have constructed a translation system to use easily even for beginners. In order to translate Japanese into Braille, we programmed in Python and C language using MeCab, a morphological analysis engine. However, there are many exceptions in Braille grammar, and the programming has become complicated. In recent years, researches on the machine translation using a neural network are thriving, so we carried out braille translation using the technology of neural machine translation (NMT) this time. Tensorflow was used as a library of neural network. We also incorporated a transform which was a translation component in the open source library Tensor2Tensor. We used 100,000 words of Japanese as teacher data. By using NMT, the morphological engine and the complicated programming become unnecessary. Also, MNT can translate without worrying about the complex Braille grammar. Then, just by entering Japanese sentence, the Braille codes are output. There are many advantages. We describe the braille translation using NMT.

By just a looking it, It’s hard for us to know if hearing impaired people have obstacles. Therefore, Hearing impaired people cannot understand of the circumstances, and they cannot get supports enough. They often live without telling they have obstacles people around them themselves. The purpose of this study, we will develop the system that visually notifies the dangerous sounds generated outside the visual field. We construct the database of environmental sounds, and select the dangerous sound from the environmental sound. This system checks the waveform data and dangerous sound database, and displays danger on the monitor if it meets conditions such as volume and distance from the source. This system displays characters using “realistic fonts” that we have developed from before, that is the expression method expressing the feeling of sound using manga technique. In this study, we will report outline of this research and a prototype of our system. We record the sound supposed to be dangerous sounds with variety of directions and range from sound source using microphone array. And we compare the waveform data for each microphone recorded, and verify the features. This system identifies three kinds of sounds and displays using realistic fonts reflecting volume and direction of sound source.

The Braille book gradually causes from ordinary printed books. Because the Braille is a reading letter and visually impaired people touch the boss with the finger’s belly, in Braille books that are frequently read, the Braille are dirty with aged use, holes open, and they collapse. In our project, we convert Braille books into machine-readable electronic data and finally restore the book. First, a Braille book is scanned by an image. Next, we detect braille by image recognition technology. And next we interpret the image, classify and identify Braille. Furthermore, we correct errors such as misdetection of braille and misidentification. Finally, we save the result as a character code. In this paper, we report techniques of elimination of gradient in the scanning images of Braille book, those of detection and identification of the Braille dot, and those of interpretation of braille dots as cells.

When a blind student enters college, many Braille books such as textbooks and reference books are needed. Students are hoping to get them before class starts. Volunteers want to give Braille books to blind students as soon as possible. However, Braille has its own rules. In translating into Braille some mistakes will occur. In order to minimize the errors, the following system is required. In other words, it is a system that converts printed textbooks into machine-readable codes, rewrites them into Kana characters in Braille rules, makes a split, corrects braille conversion mistakes and grammatical mistakes, and translates them into Braille. I will describe Japanese complexity and Japanese Braille here. I will also describe the complexity of rules when expressing Japanese Braille and the system we build.

We construct a system for a visually impaired person. On our system, an image is input by a camera, is transformed by an image conversion system, and is provided to the visually impaired person by an eyewear like glasses. If this system is used, a person can move with the eyewear, and also can watch a scene in the position of line of sight. In the eyewear composed of our system, which is made by QDLaser Inc., the image is projected directly onto a retina. QDLaser Inc. has worked on a small-sized scanning optics to be installed inside the glass frame, and has developed the eyewear named Retinal Imaging Laser Eyewear with the trademark of RETISSA. One of remarkable features of RETISSA is focus free. Regardless of the function of a crystalline lens and eyesight, the wearer can see laser scanned full-color image captured by a digital camera installed in the center of the frame connected to a controller. One of other features of RETISSA is an attractive appearance of eyewear. RETISSA also enables universal design like sunglasses and corrective glasses and realizes complete AR. We explain the system and this eyewear.

Braille books follow a process of deterioration different from ordinary printed books. Braille is a reading letter and visually impaired people touch the boss with the finger's belly. Therefore, in Braille books that are frequently read, Braille is dirty with aged use, holes open, and they collapse. In old Braille books, braille collapses and deteriorates due to pressure from left and right books in the library. Therefore, in this research, we convert Braille books into machine-readable electronic data. First, a Braille book is scanned by an image. Next, we detect braille by image recognition technology one by one from braille page. And we classify and identify Braille. Furthermore, we correct errors such as misdetection of braille and misidentification. Finally, we save the result as character code.

Generally, a hearing impaired person is supported by staffs to take a note while hearing a lecture. However, the lecture note cannot express a tone of the teacher’s voice. Further, non-verbal information such as a chatting voice in a classroom, speed, loudness and tone of speaker’s voice are also difficult to express. As a result, it is difficult for a hearing impaired person to feel the atmosphere in the classroom. In this study, we develop a system to inform atmosphere in the classroom to a hearing impaired person. The system utilizes expression techniques used in Japanese cartoons; they are “Ambient Font”, “Balloon Symbols” and “Onomatopoeic Word”. These techniques enable us to inform to the hearing impaired person not only the textual information but also the non-verbal information.

Generally, in universities it is ordinary that a small number of experts of the note taking support the hearing impaired because of the difficulty of taking a note simultaneously with hearing a lecture. The purpose of this research project is to construct a new note taking system, namely, the note taking system based upon "Quantity Rather Than Quality" concept. The origin of the name of our system is that in our system many beginners of the note taking, who is an alternative way of a skilled note taker, type the text data and finally reproduce the lecture note with the similar quality as experts do. In this study, we briefly present the mathematical model and the result of computer simulations of our note taking system. Finally, We discuss the load of the executions of voice recognitions and the feasibility of our system.

From the viewpoint of welfare costs, it is necessary to survey whether older people eat, drink and live their daily lives in safety. It is especially desired that an older person drinks green tea and/or coffee routinely and autonomously takes meals, namely breakfast, lunch and supper. In addition, it is better for an older person to eat fruits such as oranges and apples to prevent colds, Alzheimer's disease and other illnesses. In this study, a system to recognize the eating of a meal, the drinking of a beverage and the eating of a fruit was constructed using odor sensors. Three metal oxide odor sensors were adopted for this system. The system can identify behavior by analyzing characteristics of odor sensors. As for the results, the system recognizes this behavior successfully and also detects the level of fatigue of an older person. A high correlation coefficient (over 0.85) was obtained between the degree of fatigue due to bodily sensations and the output of the sensor. The sensor was highly sensitivity to hydrogen gas. When humans are fatigued or stressed, their exhaled breath contains hydrogen at high concentrations. Discriminating analysis was used to identify the kind of fruits eaten and teeth brushing behavior from odor sensor outputs.

The objective of this study is to simulate physical adsorption on the surface of semiconductor sensors and to develop a method of analyzing qualitatively and quantitatively the chemical species reacting on the surface of semiconductor sensors. In ordinary studies of the detection of gas species by semiconductor sensors, researchers attach importance to the chemical sensitivity of sensors and note the output of sensors in the stationary state in terms of response characteristics. However, the response in the transient state, the region between the beginning of adsorption and the achievement of equilibrium, contains much information about processes of gases adsorbing oil the surface of sensors. In this study, we discuss the relationship between the response of a semiconductor sensor and the amount of physically adsorbed gas and construct mathematical models of adsorption on the surface of the sensor. We simulate the signal of a model sensor using an autoregressive model in which the exponential behavior is extracted from the response, and show that gases can be identified and their concentrations can be determined.

The objective of this study is qualitative and quantitative analysis of chemical species reacting to surface of semiconductor sensors. In ordinal studies on detection of gas species by semiconductor sensors, researchers attach importance to the chemical sensitivity of sensors and notice the output value of sensors in the stationary state of the response characteristics. However, the response in the transient state, i.e., the region between the beginning of reaction and the equilibrium of reaction, has much information of chemical processes of gases reacting to surface of sensor. Hence, in this study, we adopt an autoregressive (AR) model in which the parameters describing exponential curves are looked for. As a procedure, we construct a model of reaction on the surface of sensor and assume it as the first order reaction in gas molecule. We simulate the chemical processes mathematically, and apply the AR model on the signal. Then, we reproduce the reaction rate constants and the initial amplitudes and identify the species of gases.

Three types of metal oxide gas sensors were adopted to detect the degree of breath odor. Various sorts of information are included in the odor. Each sensor has different sensitivities to gaseous chemical substances and the sensitivities also differ according to human behaviors, for example taking a meal, teeth-brushing and drinking something. There is also a possibility that the sensor can detect degrees of daily fatigue. Sensor sensitivities were low for the expiration of the elderly when the subject drank green tea. In this study, it is thought that the odor system can be incorporated into a healing robot. The robot can communicate with the elderly using several words and also connect to Internet. As for the results, the robot can identify basic human behaviors and recognize the living conditions of the resident. Moreover, it can also execute a kind of information retrieval through the Internet. Therefore, it has healing capability for the aged, and can also receive and transmit information.

This chapter focuses on synthesis of aspulvinone dimethylallyltransferase from mycelia of Aspergiilus terreus. The enzyme catalyzes the transfer of the dimethylallyl moiety from dimethylallyl pyrophosphate to activated positions ortho to the hydroxyl group of the two aromatic rings of aspulvinone E to give aspulvinone I (monoprenyl derivative) and aspulvinone H (diprenyl derivative) both of which are found as metabolites in A. terreus. The enzyme also catalyzes the prenylation of aspulvinone G, which is another metabolite of this series of pigments to give the corresponding prenylated product, aspulvinone J, which has not been found in the fungi but is assumed to be a precursor for aspulvinone C, D, and F. The principle and procedure of the assay method is described in the chapter. The prenylated product synthesized from aspulvinone E and dimethylallyl pyrophosphate, either one of which is radiolabeled, in the presence of enzyme is extracted with ether and assayed for radioactivity. Neither, aspulvinone E nor dimethylallyl pyrophosphate is extractable at neutral pH with ether. Aspulvinone E can be replaced by aspulvinone G which is also prenylated under the same conditions.

Chemischer InformationsdienstVolume 12, Issue 40 Other Subjects ChemInform Abstract: ENZYMIC SYNTHESIS OF THE INSECT JUVENILE HORMONE SKELETON T. KOYAMA, T. KOYAMASearch for more papers by this authorK. OGURA, K. OGURASearch for more papers by this authorS. SETO, S. SETOSearch for more papers by this author T. KOYAMA, T. KOYAMASearch for more papers by this authorK. OGURA, K. OGURASearch for more papers by this authorS. SETO, S. SETOSearch for more papers by this author First published: October 6, 1981 https://doi.org/10.1002/chin.198140351AboutPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat No abstract is available for this article. Volume12, Issue40October 6, 1981 RelatedInformation

AbstractEnzymatische Kondensation des Homogeranylpyrophosphats (I) mit dem Methylpentenylpyrophosphat (IIa) bzw. seinem Z‐Isomeren (IIb) und nachfolgende Hydrolyse mit alkalischer Phosphatase geben die Alkohole (IIIa) bzw. (IIIb); die daraus erhältlichen Aldehyde (IVa) bzw. (IVb) liefern durch selektive Reduktion der 2,3‐Doppelbindung die isomeren 1′7Ä Faranale (Va) + (VIa) bzw. (Vb) + (VIb) (jeweils trennbar).