Hosei University Repository
Not a member yet
22008 research outputs found
Sort by
A Case Study of Water Environment Improvement Projects in Developing Countries : Yamaha Motor Clean Water System
Get PDFThe development process of YAMAHA Motor BOP water business is explored in this paper. In the mid-1990s, Yamaha started a new research and development project to make a water-purification system for undeveloped villages. The company wanted to develop a system that could clean the surface water from rivers and lakes. In 2000, Yamaha Motor developed the Yamaha Clean Water Supply System (YCWSS) in Indonesia. As a result of this pilot plant, Yamaha Motor was able to develop a water purification system that uses a slow sand filtration method to purify water. This method is similar to the way water is purified in nature. The system has a simple design and is very compact. Also, it does not require annual filter replacement or maintenance by specialized technicians. In addition, the system requires moderate amounts of electricity to operate, making it possible to power the system with solar panels
Research on Real-time Voice Adaptation with Speech Features as Interpreted Objectives
Get PDFIn the field of speech processing, Voice Adaptation (VA), a technique for translating a spoken message from a source voice into a target voice while retaining the prosodic information, has seen serious attempts since 1980s. In order to successfully swap the spoken message between voices, the human voice which is a very complex and variable set of information must be comparable between two speakers uttering the same phrase. To do this, early research saw the use of probabilistic methods such as Gaussian Mixture Models, pitch shifting and complex mathematical models to represent a human’s physical traits such as gender, weight, vocal tract, tongue and more. In order to train the model, many speech samples, often from multiple sources were used just to be able to utter a few letters or words. In the 2010s the emergence of Deep Neural Networks (DNN) saw many uses, one being voice adaptation. While results were better than classical methods, DNN is a complex procedure that require a lot of data and time to process it in order to create a model. The more it trains, the better the model becomes; thus, it is not ideal for dynamic environments where information must be changed or added quickly. While historically, adaptation quality and amount has improved, it is still rated near the average mark quality-wise when objective evaluations are performed, and the technology is not mature enough to see serious real-world use. One area of use that is especially interesting is online in video games. In a setting where it’s commonplace to mask oneself with a new name and visual avatar, except for the voice, this is a perfect area to introduce a lightweight working VA method into.We propose a new design for training and performing VA that takes advantage of evolutionary computing to find the ideal match for a comparing function. Rather than considering language, grammar, and physique, we consider the voice as a collection of sounds a person can make. This is determined by phonemes, which vary by language, but in essence are a set of determined sounds that are used to pronounce any word for the current language. Thus, if we can collect these phonemes in conjunction with varying parameters, it should be possible to puzzle together a desired output based on the structure of the input sound. First, a short recording is made based on a manuscript reading which is designed to make the voice subject utter as many phonemes and word-combinations as possible. This audio is split into tiny fragments referred to as frames. Each frame is put through a stylized form of quantization which is a normalization stage. The processed frame represents an abstraction of the original acoustic information, which can make it easier to compare similar utterances. At the same time, other temporal and spectral features are extracted, and a selection of most crucial features are encoded into two separate color objects with RGBA-channels to group the temporal and spectral features. This information along with where they link to in the original audio is stored as a voice profile. During the actual adaptation, similar splitting and quantization are performed on the source speaker’s voice. Their color-encoded frames are used in a multi-objective optimization problem which while solving an objective function evolves over generations to present an increasingly more ideal set of target solutions. The objective function is a hybrid of a native benchmark problem and a custom function which measures the distance, or likeness between two colors, namely the source color and a target color obtained from the voice profile. Technically this is a 2-objective evaluation, but due to how the various features are encoded into color channels and the color likeness is measured, 6 objectives are being evaluated at once. When the evolution is stopped, we can with confidence select one of the solutions as an ideal target frame and put it through final adjustments before outputted. We evaluate this research using standard objective evaluation methods such as Mean Opinion Score (MOS) and ABX, MUltiple Stimuli with Hidden Reference and Anchor (MUSHRA) to evaluate the sound quality and consistency, and Single Ease Question (SEQ) to survey the interest in VA for games as well as benchmarking the performance. We see that the latest proposed iteration that has many vocal features encoded into colors for the multi-objective optimization problem achieves better quality and the training time is remarkable compared to traditional DNN.To conclude, the proposed VA system proves that it smaller amounts of speech data is sufficient in order to create a working VA as long as it is obtained through predetermined parameters such as speech intensity and phoneme content, and that multi-objective optimization problems can contribute to finding a more ideal target frame than arbitrary rules and subjective assumptions, which makes this suitable to see use in an interactive, real-time software such as games where anyone can attempt to use it.博士(理学)法政大学 (Hosei University
Structure and Limitations of Crisis Management Policies : The Great East Japan Earthquake as a Case Study
Get PDF博士(公共政策学)法政大学 (Hosei University
Research on the formation of the normative principle of environmental accounting : Explore from coexistence with nature
Get PDF博士(サステイナビリティ学)法政大学 (Hosei University
Comprehensive Research on Takahashi Gozan : Design, Illustrated Children\u27s Magazines, and Kamishibai
Get PDF博士(学術)法政大学 (Hosei University
“History of Hosei University” in Pre-War : From the Meiji 30\u27s to the End of the War
Get PDF本稿では戦前期法政大学において、法政大学の歴史について述べた各種資料を紹介した。法政大学の歴史を述べた最初期の資料は明治30年代に見出すことができ、これらの資料として、富井政章（和仏法律学校校長）の演説、『法学志林』に掲載された「私立法政大学一覧－沿革略」を挙げることができる。「沿革略」に記された、東京法学社（法政大学の起源）の明治12年創立説は1950年代後半の『法政大学八十年史』（1961年）で訂正されるまで、法政大学史の定説であった。1909（明治42）年には、法政大学史としては初の出版物である石原三郎編『法政大学参拾年史』が刊行された。同著は創立30周年記念の一環として編纂されたもので、石原ら大学の事務職員が編纂にあたり、戦前期の法政大学史を代表する著作である。1928（昭和3）年に実施された創立50周年事業では、法政大学の歴史に関する各種講演が行われ、これらの講演でも紹介されていた明治期の学生生活に関する回想などが、以後、学内誌に掲載されるようになる。1941（昭和16）年には、「法政大学六十年史」が『法政大学校友名鑑』に掲載された。「六十年史」は「小史」とも呼ぶべきものであり、その内容は組織史・制度史を中心としているが、明治30年代に法政大学は設置した清国留学生法政速成科について述べた部分などは戦時下という時局の影響を受けた記述となっている。In this article, the author listed a range of materials on the history of Hosei University in pre-war. The oldest materials describing the history of the University are those written during the Meiji 30’s. These include a speech on the brief history of the University made by Masaakira Tomii, president of Wafutsu Horitsu Gakko (School of Japanese and French Law) and published in Hogaku Shirin. The brief history stated that the University was originally founded in 1879, which continued to be widely believed until the year was corrected in the publication on Hosei University’s 80-year history, created in the latter half of the 1950s and published in 1961.In 1909, the first publication on Hosei University’s history was issued as part of the project to celebrate the 30th anniversary of its founding, with Saburo Ishihara and other staff from the University serving as its editors. It became one of the most prominent representative publications on the pre-war history of Hosei University.In the project implemented in 1928 to celebrate the 50th anniversary of its founding, a range of lectures was given on the University’s history, and campus life in the Meiji period, which was also mentioned in the lectures, began to be introduced in the University’s in-house magazine.In 1941, the 60-year history of Hosei University was described in the Hosei University alumni directory. The document summarized the organizational and institutional history of the University. Some parts of the document were written under the influence of the war, including the part on the intensive course for students from Qing China, which the University provided during the Meiji era
EFFECT OF MEMBRANE FOULING ON SOLUTE REMOVAL PERFORMANCE OF DIALYZERS WITH DIFFERENT MEMBRANE AREA
Get PDFIn blood purification therapy, the membrane fouling due to protein molecules reduces solute removal performance. We investigated the relationship between the membrane fouling and membrane area of dialyzer systematically, by reproducing the fouling in an aqueous system. The decreasing degree of solute removal performances depended on the membrane area of dialyzer, and a higher clearance was kept in dialyzers with larger membrane area. Increased effect of the internal filtration, filtration from the inside to outside and its opposite occurring in one dialyzer at the same time, found with increased membrane area enhanced the removal of middle molecules regardless of the membrane fouling
EFFECT OF INTERMITTENT BACK FILTRATION ON SOLUTE REMOVAL IN HEMODIAFILTRATION
Get PDFAn intermittent infusion of perfectly purified dialysate into the blood compartment across the membrane using back filtration in blood purification therapy induces a membrane flushing during treatment. We devised a method to flush the membrane effectively and tried to find the ideal back filtration patterns for the purpose of solute removal based on the results of computing the pressure distribution in an artificial kidney. It was found that the solute removal performance of inulin remained at the highest level when the back filtration was performed under parallel flow
PHOTOLUMINESCENCE CONTROL OF COLLOIDAL SILICON NANOCRYSTALS BY CHEMICAL ETCHING OF RAW POROUS SILICON POWDER
Get PDFWe demonstrated light emission color control of colloidal silicon nanocrystals by chemical etching of porous silicon (PSi) flakes. The colloidal silicon nanocrystals derived from post-etched PSi flakes in hydrofluoric acid (HF) and ethanol show different light emission color due to reduction of silicon core size. The colors are changed from red to yellow depending on the etching conditions such as time and composition of etching