ゼミナール発表

日時: 9月24日(木)2限 (11:00-12:30)


会場: L1

司会: 松原 崇充
小林 和弘 1461002: D, 中間発表 中村 哲,杉本 謙二,戸田 智基,Sakriani Sakti,Graham Neubig
title: Statistical singing voice conversion based on direct waveform modification
abstract: In this talk, I will speak statistical singing voice conversion (SVC) technique with direct waveform modification based on the spectrum differential. In prior research, several SVC techniques make it possible to get rid of physical constraints of singers by expanding their singing expression such as converting voice timbre. However, sound quality of the converted singing voice is significantly degraded compared to that of a natural singing voice due to various factors, such as analysis and modeling errors in the vocoder-based framework. To alleviate this degradation, I have proposed a statistical conversion process that directly modifies the signal in the waveform domain by estimating the difference in the spectra of the source and target singing voices. The differential spectral feature is directly estimated using a differential Gaussian mixture model (GMM) that is analytically derived from the traditional GMM used as a conversion model in the conventional SVC frameworks. The experimental results demonstrate that the proposed method makes it possible to significantly improve sound quality in the converted singing voice while preserving the conversion accuracy of singer identity compared to the conventional SVC.
language of the presentation: Japanese
 
田尻 祐介 1451069: M, 2回目発表 中村 哲,杉本 謙二,戸田 智基,Sakriani Sakti,Graham Neubig
title: Noise suppression method for body-conducted soft speech based on external noise monitoring
abstract: Recently, silent speech interfaces have attracted attention as a technology that enables us to talk with each other while not being overheard by others. Non-Audible Murmur (NAM) microphone was developed to detect an extremely soft speech through the body of speaker and it is effectively used as such a silent speech interface. Although the NAM microphone is more robust against external noise compared to an air-conductive microphone, body-conducted soft speech signals will still suffer from them due to its faint volume. To address this issue, we propose a noise suppression method based on external noise monitoring.
language of the presentation: Japanese
 
倶羅 真也 1451043: M, 2回目発表 中村 哲,杉本 謙二,戸田 智基,Sakriani Sakti,Graham Neubig
title: Duration Correction for Non-Native Speech Based on Waveform Processing
abstract: Japanese and English differ pronunciation, intonation and rhythm. Therefore native English uttered by Japanese deteriorates the naturalness of the communication. This study proposes the naturalness improvement method based on the duration modification of English uttered by Japanese. Although the conventional method performs the correction using vocoder, the speech deteriorates the naturalness of the modified speech. The proposed method modifies based on waveform processing for non-native speech. This talk presents the result of the duration correction of the native English speech and task.
language of the presentation: Japanese
発表題目: 音声波形加工に基づく非母語話者音声の継続長補正
発表概要: 日本語と英語は発音,イントネーション,リズムが大きく異なる.そのため,日本語母語話者は英語を発話する時,それらの違いから不自然な音声を発話する傾向がある.そこで本研究では,発話様式の違いの一つである発話リズムの補正により,日本人英語の自然性を改善する手法を提案する.従来は,発話リズムの補正の際にボコーダーを用いており,補正後の音質が劣化していた.そこで,非母語話者の音声波形に対して直接補正処理を行う手法を提案する.本報告では,実際に発話リズムを補正した結果を示し,そこから得られた課題について報告する.
 
高山 怜也 1451065: M, 2回目発表 中村 哲,向川 康博,戸田 智基,Sakriani Sakti,Graham Neubig
title: Mental illness detection with linguistic feature on social media
abstract: Recently, there are growing interests in posts on social media, due to the concern over the big data has risen. In social media, some users who have various background post their opinion or experience and this equal to their life log. We analyze those data with linguistic features and aim to find mental illness in lower cost by comparison to going hospital. On this presentation, we will talk about present condition, plan of collecting data and future analysis.
language of the presentation: Japanese
 

会場: L2

司会: 佐藤 哲大
岩本 昌大 1451016: M, 2回目発表 小笠原 司,萩田 紀博,高松 淳,吉川 雅博
title: Development of Android to Tell a Lie Based on Behavioral Psychology
abstract: In this study, by develop a Android to tell a lie based on behavioral psychology, we aim to achieve a more human interaction between humans and Android. Achieve of human interaction, it is necessary to take a communications interspersed with lie. In this presentation, the conducted experiments to verify the implementation lie gesture, and consider the result.
language of the presentation: Japanese
発表題目: 行動心理学に基づいた嘘を付くアンドロイドの開発
発表概要: 本研究では、行動心理学に基づいた嘘を付くアンドロイドを開発することで、人間とアンドロイドとのより人間らしい対話の実現を目指す。人間らしい対話の実現は、嘘を交えたコミュニケーションを取る事が必要である。本発表では実装した嘘仕草について検証する実験をし、その結果を考察する。
 
松永 拓也 1451099: M, 2回目発表 小笠原 司,金谷 重彦,高松 淳,吉川 雅博
title: Daily life assistive hand for finger dysfunction
abstract: Functional grasping is required in activities of daily living (ADLs). However, the grasing ability of individuals may decrease in the cases of cervical spine damage or severed fingers. Hence, supporting the grasping ability in ADLs of a person with impairment of upper limbs is important. With this objective in mind, we developed an assistive hand for daily life use. The hand is mounted on the forearm with a prosthetic socket and is capable of grasping objects of daily life environments. We designed different shapes for the prosthetic socket and conducted experiments to assess user impressions in order to increase the usability of the hand.
language of the presentation: Japanese
発表題目: 手指機能障害のための日常生活支援ハンドの開発
発表概要: 日常生活の中では様々な把持を行っているが,頸椎損傷や手指欠損による手指機能障害者は把持能力が低下するため,ADLが低下している.把持動作を補助する外骨格型の支援装置が開発されているが,把持能力や装着性に問題がある.そこで,本研究では,手指機能障害者の日常生活支援を目的とし,前腕装着型のハンドを開発する.本ハンドは,ソケットを用いて前腕に装着し,様々な把持を実現することが可能である.その中でも,ソケットの形状やハンドと手の位置関係は,開発するハンドの使用感を左右する重要な要素だと考えられる.本発表では,開発したハンドの構成や仕様について説明し,ソケットの形状に関して行った実験について考察する.
 
向山 寛人 1451107: M, 2回目発表 小笠原 司,佐藤 嘉伸,高松 淳,丁 明
title:Contact force estimation using a finger deformation measurement sensor
abstract:we propose both a novel fingertip lateral deformation (FLD) sensor and an estimation method of fingertip force using a system identification technique. A transfer function model (multi-input multi-output) is employed for the normal force, the tangential force estimation from FLD sensor data. We show the experimental results to consider the influence of the contact conditions such as the normal force, motion speed and friction to the physical contact parameter estimation. Finally, we propose the required calibration conditions of physical contact parameter estimation using FLD sensor based on the experimental results.
language of the presentation: Japanese
発表題目: 指腹部変形計測センサを用いた接触力推定
発表概要: 物体把持および操作における指先の指腹部変形に着目し,指腹部変形から指先にかかっている力と滑りの推定を行うことを目的とする.自然な接触を妨げない計測センサを利用して指腹部変形を計測し,指腹部変形と指先にかかっている力の関係をモデル化することで,指先力の推定を行う.また,接触力や指先の移動速度,摩擦係数がモデルのパラメータ推定に与える影響を考察する.
 
近藤 雅也 1451202: M, 2回目発表 小笠原 司,佐藤 嘉伸,多田 充徳,高松 淳
title: Musculoskeletal Hand Model with Individual Differences
abstract: In this study, we developed a subject-specific musculoskeletal hand model considering the differences in inertia parameters and muscle force parameters. Furthermore, we verified the estimated maximum pinching force compared with measured sensor force data while subjects were pinching 28 mm wide object by using developed model. As a result of the simulation, it is suggested that it is possible to perform musculoskeletal simulation considering individual differences.
language of the presentation: Japanese
発表題目: 個人別手部筋骨格モデルの構築
発表概要: 本研究では,慣性特性や筋特性などの個人差を考慮した被験者特有の手部筋骨格モデルを構築した.また構築したモデルを用いて,幅が28[mm]の物体のつまみ動作時における最大つまみ力の推定および検証を行った結果,個人差を考慮した筋骨格シミュレーションが行える可能性が示唆された.
 

会場: L3

司会: 藤本 まなと
CRUZ JASON PAUL MIRANDA 1461014: D, 中間発表 伊藤 実,藤川 和利,楫 勇一,関 浩之
title: The Bitcoin Network as Platform for Trans-Organizational Attribute
abstract: The role-based access control (RBAC) is a natural and versatile model of the access control principle. In the real world, it is common that an organization provides a service to a user who owns a certain role that was issued by a different organization. However, such a trans-organizational RBAC is not common in a computer network because it is difficult to establish both the security that prohibits malicious impersonation of roles and the flexibility that allows small organizations/individual users to fully control their own roles. This study proposes a system that makes use of Bitcoin technology to realize a trans-organizational RBAC mechanism. Bitcoin, the first decentralized digital currency, is a payment network that has become a platform for innovative ideas. Bitcoin's technology, including its protocol, cryptography, and open-source nature, has built a good reputation and has been applied in other applications, such as trusted timestamping. The proposed system uses Bitcoin technology as a versatile infrastructure to represent the trust and endorsement relationship that are essential in RBAC and to realize a challenge-response authentication protocol that verifies a user's ownership of roles.
language of the presentation: English
 
武田 友希 1451068: M, 2回目発表 伊藤 実,安本 慶一,楫 勇一
title: On the Computational Complexity of the Solvability of Information Flow Problem with Hierarchy Constraint
abstract: An information flow problem discusses how to distribute information over a complicated network. It is known that the technique of the network coding plays an essential role in a certain type of information flow problems, but not so much are known about other types of the problem. As an extension of Lehman's investigation, this study introduces a hierarchy constraint of messages, and discusses the computational complexity of the problem to determine if a given information flow problem has a linear solution or not. Nine classes of problems are newly defined, and classified to one of three categories that were discovered by Lehman.
language of the presentation: Japanese
発表題目: 階層制約のある情報フロー問題に関する計算理論的考察
発表概要: 情報フロー問題は,情報の持つ特性を考慮した上で,ネットワーク上で情報を流通させる方式を検討する問題である.ある種の情報フロー問題では,線形ネットワーク符号が重要な役割を果たすことが知られているが,一般的な問題における線形ネットワーク符号の特性については明らかでない部分も多い.本研究では,与えられた情報フロー問題が線形解を持つか否か判定する問題において,Lehman らによる既存研究を拡張し,メッセージ集合に階層制約のある情報フロー問題について計算理論的な考察を行う.階層制約は,現実のネットワークサービスの中でも頻繁に発生する事例であり,この制約下の情報フロー問題を考えることは理論的に興味深いだけでなく,実用的観点からも重要である.他の制約と組み合わせて新しく導入される9種類の情報フロー問題それぞれが,Lehman の3カテゴリのいずれに分類されるかを明らかにする.
 
宵 憲治 1451119: M, 2回目発表 安本 慶一,伊藤 実,荒川 豊,諏訪 博彦,藤本 まなと
title: An Appliance Control System for Reaching Target Temperature and Humidity with Minimal Cost in the room abstract: Aiming to realize a context-aware system for controlling various home appliances in a home, we propose an appliance control system for reaching target temperature and humidity with minimal energy cost. The proposed system changes the temperature and humidity of the room to the target values by using a tool called PathSim developed in our previous research which derives an appliance operation sequence for moving between two any context at the minimum cost through simulation. In this presentation, we describe the overview of PathSim and the architecture of the proposed system. Moreover we provide the results and discussion of our experiment in a smart home and show the effectiveness of the proposed system. language of the presentation:Japanese 発表題目: 室内温湿度環境の遷移を最小コストで実現する家電制御システム 発表概要: スマートホームにおいて様々な家電デバイスを制御するコンテキストア ウェアシステムの実現を目的に,スマートホームにおいて目標温湿度環境への遷移を最小コストで実現する家電制御システムを提案する.提案システムは,著者らが以前から開発してきた任意の2つのコンテキスト間を最小コストで遷移させる家電操作系列をシミュレーションにより求めるツールPathSimを用いて,実空間における温湿度を目標の値に最小コストで 遷移させる.本発表では,PathSimの 概要および提案システムのアーキテクチャを述べ,その後システムの有用性を示すため実際のスマートホーム環境で行った実証実験の結果と考察,今後の課題について述べる.
 
竹内 昌平 1451067: M, 2回目発表 中島 康彦,伊藤 実,高前田 伸也,TRAN THI HONG
title:A Distributed Memory Based Embedded CGRA for Accelerating Stencil Computations
abstract:Energy efficient, small footprint, and high performance accelerator is necessary in embedded system. I developed an embedded CGRA accelerator with distributed on-chip memory blocks for both energy- and memory-bandwidth-efficient stencil computation. In this research, I implemented a real LSI and its FPGA based evaluation platform by using Xilinx Zynq and Debian Linux. The evaluation result shows that the accelerator achieves higher performance and lower energy consumption, compared to ARM core with Zynq. Furthermore, estimation result shows that the accelerator manufactured in 28nm process achieves 1.61x better energy efficiency than the mobile GPU.
language of the presentation:Japanese
発表題目:組み込みシステム向けメモリ分散型CGRAの性能評価
発表概要:組み込みシステムでは、高い電力効率、小回路面積かつ高性能な計算プラットフォームが求められている。そこで、我々はCGRA型の組み込みシステム向けアクセラレータを開発した。ステンシル計算プログラムにおいて、このアクセラレータは内部に分散配置されたメモリを活用することで低消費電力を達成する。本研究では、試作チップとZynqを用いてCPUとアクセラレータがメインメモリを共有するシステムを実装した。評価では、EMAXはZynq単体と比較して高い性能と低い消費電力量を達成した。さらに、28nmプロセスでEMAXを作製した場合、小面積のGPUと比較して最大で1.61倍の性能を見積もった。