|PATRICK LUMBANTOBING||1451207: M, 2回目発表||知能コミュニケーション 中村 哲, 杉本 謙二, 戸田 智基, Sakriani Sakti, Graham Neubig|
title: Articulatory Controllable Speech Modification based on a Sequential Statistical Feature Mapping System with Gaussian Mixture Model
abstract: The movements of articulators, e.g. tongue and lips, define the resonance characteristics of the vocal tract during a speech production process. Therefore, the spectrum of the vocal tract can be parameterized not only by acoustic parameters, such as mel-cepstrum, but also by the more slowly varying articulatory parameters. In this work, in order to take advantage of the articulatory parameters' characteristics, an intuitive speech modification system is developed by making it possible to modify an input speech signal through the manipulation of the unobserved articulatory parameters. Given the acoustic spectrum of the input speech signal, the corresponding articulatory parameters are estimated with a statistical inversion mapping system using a Gaussian mixture model. The estimated articulatory parameters can be manually manipulated by also taking into account the correlation between articulatory organs. The modified acoustic spectrum is then estimated from the manipulated articulatory movements with a statistical production mapping system using a Gaussian mixture model. The experimental results demonstrated that 1) the proposed system is capable of generating high-quality modified speech sounds in certain speaking conditions, e.g. hypo- and hyper-articulated speech, and 2) it is possible to modify pronunciation of particular phonemic sound into another by manipulating the corresponding articulatory positions.
language of the presentation: English
|野田 哲男||1561202: D, 中間発表||ロボティクス 小笠原 司, 杉本 謙二, 高松 淳|
title: Study on the Intelligentization of the Industrial Robot for Manufacturing by the Autonomous System Algorithm |
abstract: There are two impediments to widespread adoption of the industrial robot system in the manufacturing field. One is due to the engineering cost of production system construction. Another is the presence of a work that is difficult for robots though it is easily done by human hand. Human who is using the autonomy (spontaneous ingenuity) have overcome these challenges naturally. In this study, the presenter has proposed his autonomous system algorithms solving them, and has reported their effects in the demonstration system. The first is The Active Search Algorithm to solve the optimization problem of unknown objective function. The second is The Re-Grip Algorithm to solve the random bin picking problem for bulk parts feeding that is always a nuisance of fields.
language of the presentation: Japanese
発表概要: 製造業向けロボットシステムの幅広い普及の阻害要因は２種ある．それらは，(1)生産システム構築のエンジニアリングコストに起因するもの，(2)人手では容易だが，ロボット化が困難な作業の存在，である．人は自律性(自発的な創意工夫)で，これらの難題を自然に克服している． 本研究では，それら問題の困難性の本質を解決してロボットを知能化する自律システムアルゴリズムを提案し，実証システムにおいてその効果を確認したので報告する． ひとつめは，未知の目的関数の最適化問題を解決する能型探索アルゴリズムである． ふたつめは，現場で必ず問題となるバラ積み部品供給におけるランダム・ビン・ピッキング問題を解決する持ち替えアルゴリズムである．
|NGUYEN THE TUNG||1551205: M, 1回目発表||知能コミュニケーション 中村 哲|
title: *** Considering deception information in negotiation dialog. ***
abstract: ***Lying is a part of daily human's activities. According to a recent survey, about 25% of people admitted that they doesn't tell the truth when talking with their doctors. This makes it very difficult for the doctor to obtain necessary information to formulate a diagnosis and provide medical treatment. Thus, I introduced a spoken dialog system that can perform deception detection using both verbal and non-verbal information such as facial expressions from the user. By incorporating a deception detection module into the sysetm and use this information in the dialog state tracking step, we expect to improve the state tracking accuracy and increase the overall performance of the system. The dialog system utilize Gaussian process POMDP for dialog management. ***
language of the presentation: English