Updated on 2026/03/10

写真a

 
SHINOZAKI TAKAHIRO
 
Organization
School of Engineering Professor
Title
Professor
External link

Degree

  • 博士(学術) ( 2004.3 )

Research Interests

  • Automatic speech recognition

  • pattern recognition

  • statistical model

Research Areas

  • Informatics / Intelligent robotics

Education

  • Tokyo Institute of Technology   Graduate School of Information Science and Engineering   Department of Computer Science

    - 2004

      More details

    Country: Japan

    researchmap

Research History

  • Institute of Science Tokyo

    2024.7

      More details

  • Tokyo Institute of Technology   Associate Professor

    2016.4 - 2024.6

      More details

  • Tokyo Institute of Technology   Associate Professor

    2013.3 - 2016.3

      More details

  • Chiba University   Assistant Professor

    2011.4 - 2013.2

      More details

  • Tokyo Institute of Technology   Department of Computer Science   Assistant Professor

    2008.10 - 2011.3

      More details

  • :Tokyo Institute of Technology   Graduate School of Information Science and Engineering   Research Fellow

    2007 - 2008

      More details

  • :Kyoto University   Academic Center for Computing and Media Studies   Research Assistant Professor

    2006 - 2007

      More details

  • :University of Washington   Department of Electrical Engineering   Research Scholar

    2004 - 2006

      More details

▼display all

Professional Memberships

Committee Memberships

  • 音響学会   音声研究会 主査  

    2025   

      More details

  • 情報処理学会/電子情報通信学会   音声言語情報処理研究会/音声研究会 主査  

    2024   

      More details

  • 日本学術会議   計算音響学小委員会  

    2021.2   

      More details

    Committee type:Government

    researchmap

  • 情報処理学会   JIP編集委員  

    2020.6   

      More details

    Committee type:Academic society

    researchmap

  • 電子情報通信学会   ISS誌編集委員(SP担当)  

    2012.6   

      More details

    Committee type:Academic society

    researchmap

Papers

▼display all

MISC

  • 超多言語事前学習による低資源音声認識の検討

    Hou Wenxin, Dong Yue, ZHUANG BAIRONG, 楊 龍飛, 篠崎隆宏

    日本音響学会   ( 2-P1-7 )   2020.9

     More details

    Language:Japanese  

    researchmap

  • Transformer 音声認識システムの進化的最適化

    日野 健人, 篠崎隆宏

    日本音響学会2020年秋季研究発表会講演論文集   2-P1-6   2020.9

     More details

    Language:Japanese  

    researchmap

  • 二重相続進化戦略による音声認識システムの最適化

    日野 健人, 木村 友祐, Dong Yue, 篠崎隆宏

    日本音響学会2020年春季研究発表会講演論文集   2-4-5   893 - 894   2020.3

     More details

    Language:Japanese  

    researchmap

  • CNNフロントエンドによる高速なEnd-to-End連続DPマッチングの実現

    田中 智宏, 篠崎隆宏

    日本音響学会2020年春季研究発表会講演論文集   2-4-4   891 - 892   2020.3

     More details

    Language:Japanese  

    researchmap

  • Robust Multichannel End-to-End Speech Recognition Based on Multi-Output Densenet

    Chonghui Zheng, Takahiro Shinozaki

    2020-SLP-131 ( No. 10 )   1 - 3   2020.2

     More details

    Language:English  

    researchmap

  • 二重相続進化戦略によるEnd-to-End音声認識システムの最適化

    木村 友祐, 日野 健人, DongYue, 篠崎 隆宏

    研究報告音声言語情報処理(SLP)   2020-SLP-131 ( No. 11 )   1 - 3   2020.2

     More details

    Language:Japanese  

    researchmap

  • Efficient Spoken Language Acquisition Based on Learning Synergy Principle

    篠崎隆宏, GAO Shengzhou, ZHANG Mingxin, HOU Wenxin, 田中智宏

    人工知能学会言語・音声理解と対話処理研究会資料   89th   2020

  • CNNフロントエンドによるEnd-to-End連続DPマッチングの高速化

    田中 智宏, 篠崎 隆宏

    研究報告音声言語情報処理(SLP)   Vol. 2019-SLP-130 ( No. 2 )   2019.12

     More details

    Language:Japanese  

    researchmap

  • 入力画像勾配を用いたモデル構造フリーな教師無し音源ローカライゼーション

    田中 智宏, 篠崎隆宏

    日本音響学会2019年秋季研究発表会講演論文集   2-3-3   919 - 920   2019.9

     More details

    Language:Japanese  

    researchmap

  • 営業電話における大規模 End-to-End 音声認識システムの活用

    平村 健勝, 篠崎隆宏

    日本音響学会2019年秋季研究発表会講演論文集   1-3-3   1183 - 1184   2019.9

     More details

    Language:Japanese  

    researchmap

  • Aggregated CMA-ES: An Effective and Stable Strategy for Neuron Model Optimization

    Xu Han, Takahiro Shinozaki, Ryota Kobayashi

    ( No. 9 )   1 - 2   2019.3

     More details

    Language:English  

    researchmap

  • 連続単語検出のための 2D-RNN を用いた End-to-EndDPマッチング

    田中智宏, 篠崎隆宏

    日本音響学会2019年春季研究発表会講演論文集   ( 2-P-13 )   979 - 980   2019.3

     More details

    Language:Japanese  

    researchmap

  • Analysis of Attention-Based Multimodal Fusion and Maximum Mutual Information Objective for DSTC7 Audio Visual Scene-Aware Dialog Track

    Wenbo Wang, Bairong Zhuang, Takahiro Shinozaki

    ( 2-P-10 )   973 - 974   2019.3

     More details

    Language:English  

    researchmap

  • 連続対応検出ネットワークによる音声動画からの教師なし物体セグメンテーションおよび関連学習の検討

    田中智宏, 篠崎隆宏

    日本音響学会2019年春季研究発表会講演論文集   ( 2-P-13 )   979 - 980   2019.3

     More details

    Language:Japanese  

    researchmap

  • 大規模 End-to-End 音声認識システムの教師なし強化学習の実現に向けた検討

    PengYilong, 篠崎隆宏

    日本音響学会2019年春季研究発表会講演論文集   ( 1-P-9 )   919 - 920   2019.3

     More details

    Language:Japanese  

    researchmap

  • I-vector Domain Adaptation Using Cycle-Consistent Adversarial Networks for Speaker Recognition

    Yi Liu, Takahiro Shinozaki

    2019-SLP-126 ( No. 2 )   1 - 3   2019.2

     More details

    Language:English  

    researchmap

  • マルチゲートGRUユニットを用いた2D-RNNによるEnd-to-End始終端フリー単語検出

    田中智宏, 篠崎隆宏

    音声言語情報処理研究会   2018.12

     More details

    Language:Japanese  

    researchmap

  • Improving the audio visual scene-aware dialog system in DSTC7 by using attentional multimodal fusion and MMI objective

    Wenbo Wang, Bairong Zhuang, Takahiro Shinozaki

    2018.12

     More details

    Language:English  

    researchmap

  • 単語検出性能を目的関数とした単語検出器学習法の提案

    田中智宏, 篠崎隆宏

    2018年秋季研究発表会   2018.9

     More details

    Language:Japanese  

    researchmap

  • 音声認識システムの教師なし強化学習における報酬と報酬ノイズの影響の検討

    PengYilong, 柴田駿人, 篠崎隆宏

    2018年秋季研究発表会   2018.9

     More details

    Language:Japanese  

    researchmap

  • 強化学習による報酬のみを用いたend-to-end 認識システム学習

    柴田駿人, PengYilong, 篠崎隆宏

    2018年秋季研究発表会   2018.9

     More details

    Language:Japanese  

    researchmap

  • End-to-end音声認識システムの強化学習の検討

    PengYilong, 柴田駿人, 篠崎隆宏

    音声言語情報処理研究会   2018-SLP-123 ( 9 )   1 - 4   2018.7

     More details

    Language:Japanese  

    researchmap

  • Taxi Demand Prediction using Ensemble Model Based on RNNs and XGBOOST Reviewed

    Takahiro Shinozaki

    9th International Conference of Information and Communication Technology for Embedded Systems   130 - 135   2018.5

     More details

    Language:English  

    researchmap

  • 日本人英語学習者を対象とした自動英語音声認識の予備検討

    篠崎 隆宏, 加藤 拓

    CEFR-J 2018 Symposium   2018.3

     More details

    Language:Japanese  

    researchmap

  • End-to-Endニューラル対話モデルにおける単語分散表現の比較検討

    鄭 崇輝, 李 知雨, 王 文博, 庄 佰融, 篠崎 隆 宏

    2018年春季研究発表会講演論文集   2018.3

     More details

    Language:Japanese  

    researchmap

  • 音声認識仮説を用いたベイズ的半教師あり発音辞書学習の検討

    池下裕紀, 篠崎隆宏

    春季研究発表会講演論文集   2018.3

     More details

    Language:Japanese  

    researchmap

  • 方策勾配法と仮説選択に基づくDNN音声認識システムの強化学習

    加藤拓, 篠崎隆宏

    春季研究発表会講演論文集   2018.3

     More details

    Language:Japanese  

    researchmap

  • 英語学習者の発声自動評価を目的としたDNN音声認識システムの検討

    加藤 拓, 篠崎 隆宏

    情報処理学会研究報告   Vol. 2017-SLP-119 ( No. 11 )   1 - 4   2017.12

     More details

    Language:Japanese  

    researchmap

  • ベイズ推論を用いた半教師あり学習の日本語適用

    池下裕紀, 篠崎隆宏, 渡部晋治, 持橋大地, Graham Neubig

    情報処理学会研究報告   Vol. 2017-SLP-118 ( No. 3 )   1 - 4   2017.10

     More details

    Language:Japanese  

    researchmap

  • 仮説選択に基づくDNN音声認識システムの強化学習

    加藤 拓, 篠崎 隆宏

    情報処理学会研究報告   Vol. 2017-SLP-118 ( No. 4 )   1 - 5   2017.10

     More details

    Language:Japanese  

    researchmap

  • 進化的戦略を用いたDNNハードウエア音声センサの低消費電力化

    銭 博宇, 王 健, 劉 溢, 朱 凱, 篠崎 隆宏

    2017年秋季研究発表会講演論文集   131 - 132   2017.9

     More details

    Language:Japanese  

    researchmap

  • ゼロリソース言語への応用を目的としたABXテストによるDNN特徴量の検討

    柴田駿人, 加藤拓, 篠崎隆宏, 渡部晋治

    秋季研究発表会講演論文集   1 - 2   2017.9

     More details

    Language:Japanese  

    researchmap

  • 進化的戦略を用いたニューラル機械翻訳システムの自動最適化

    覃 浩, 篠崎 隆宏, Duh Kevin

    2017年秋季研究発表会講演論文集   1397 - 1398   2017.9

     More details

    Language:Japanese  

    researchmap

  • 読み上げ音声を用いたニューラルネットワークによる任意歌唱者歌声声質変換の検討

    篠崎隆宏, 小池治憲, 能勢隆, 伊藤彰則

    日本音響学会春季研究発表会講演論文集   357 - 358   2017.3

     More details

    Language:Japanese  

    researchmap

  • Highwayネットワーク言語モデルを用いた日本語話し言葉音声認識

    田中智大, 篠崎隆宏, 渡部晋治

    日本音響学会春季研究発表会講演論文集   107 - 108   2017.3

     More details

    Language:Japanese  

    researchmap

  • ベイズ的教師なし発音辞書学習のWFST実装およびサンプリングアルゴリズムの検討

    篠崎隆宏, 渡部晋治, 持橋大地, Graham Neubig

    日本音響学会春季研究発表会講演論文集   17 - 18   2017.3

     More details

    Language:Japanese  

    researchmap

  • Hardware Speech Sensor Based on Deep Neural Network Feature Extractor and Template Matching

    Yi Liu, Boyu Qian, Jian Wang, Takahiro Shinozaki

    116 ( 477 )   297 - 300   2017.3

     More details

    Language:English  

    CiNii Books

    researchmap

  • 半教師ありDNN学習を用いた日本語スピーキングテスト音声の認識

    加藤 拓, 篠崎 隆宏

    日本音響学会春季研究発表会講演論文集   93 - 94   2017.3

     More details

    Language:Japanese  

    researchmap

  • 敵対的学習を利用したニューラルネットワークに基づく任意話者声質変換の検討

    篠崎隆宏, 宮本 颯, 能勢 隆, 伊藤鈴乃介, 小池治憲, 伊藤彰則

    日本音響学会春季研究発表会講演論文集   355 - 356   2017.3

     More details

    Language:Japanese  

    researchmap

  • ChimeChallengeタスクにおけるNMFによる雑音除去の検討

    小澤 奈摘, 田中 智大, 篠崎 隆宏

    音声言語情報処理研究会(SLP)   Vol. 2017-SLP-115 ( No. 12 )   2017.2

     More details

    Language:Japanese  

    researchmap

  • 進化戦略に基づいた単語検出ハードウェアのためのDNNメタパラメータ最適化

    王 健, 銭 博宇, 劉溢, 篠崎 隆宏

    音声言語情報処理研究会(SLP)   Vol. 2017-SLP-115 ( No. 6 )   2017.2

     More details

    Language:Japanese  

    researchmap

  • 眼球動作に基づいた対話支援システムのための連続画なぞり入力手法 (音声) -- (第18回音声言語シンポジウム)

    房 福明, 篠崎 隆宏

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   116 ( 378 )   83 - 88   2016.12

     More details

    Language:Japanese   Publisher:電子情報通信学会  

    researchmap

  • 第3回Frederick Jelinek記念サマーワークショップでの教師なし発音辞書学習の取り組み

    篠崎隆宏, 渡部晋治, 持橋大地, Graham Neubig

    音声言語情報処理研究会 (SIG-SLP)   2016.12

     More details

    Language:Japanese  

    researchmap

  • 眼球動作に基づいた対話支援システムのための連続画なぞり入力手法

    房 福明, 篠崎 隆宏

    音声言語情報処理研究会(SLP)   Vol. 2016-SLP-114 ( No. 19 )   2016.12

     More details

    Language:Japanese  

    researchmap

  • 第3回Frederick Jelinek記念サマーワークショップでの教師なし発音辞書学習の取り組み (音声) -- (第18回音声言語シンポジウム)

    篠崎 隆宏, 渡部 晋治, 持橋 大地, Neubig Graham

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   116 ( 378 )   11 - 15   2016.12

     More details

    Language:Japanese   Publisher:電子情報通信学会  

    researchmap

  • 日本語話し言葉音声における半教師ありDNN学習の検討

    加藤 拓, 篠崎 隆宏

    音声言語情報処理研究会 (SIG-SLP)   Vol. 2016-SLP-113 ( No. 1 )   2016.10

     More details

    Language:Japanese  

    researchmap

  • Automatic speech recognition and black-box optimization

    72 ( 10 )   644 - 652   2016.10

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • 連続音声認識におけるLSTMによる単語履歴を考慮した未知語検出法

    池下裕紀, 篠崎隆宏

    日本音響学会秋季研究発表会   2016.9

     More details

    Language:Japanese  

    researchmap

  • 差分スペクトルフィルタに基づく声質変換における性能向上の検討

    小池治憲, 能勢 隆, 篠崎隆宏, 伊藤彰則

    日本音響学会秋季研究発表会講演論文集   285 - 286   2016.9

     More details

    Language:Japanese  

    researchmap

  • 進化的戦略を用いたリカレントニューラルネットワーク言語モデルの最適化

    田中智大, 森谷崇史, 篠崎隆宏, 渡部晋治, 堀貴明, Kevin Duh

    日本音響学会秋季研究発表会講演論文集   31 - 32   2016.9

     More details

    Language:Japanese  

    researchmap

  • LSTMによる単語履歴を考慮した未知語検出法

    池下裕紀, 篠崎隆宏

    音声研究会(SP)   116 ( 189 )   33 - 36   2016.8

     More details

    Language:Japanese   Publisher:電子情報通信学会  

    CiNii Books

    researchmap

  • 国際会議ICASSP2016参加報告

    峯松信明, 秋田祐哉, 浅見太一, 伊藤信貴, 落合翼, 郡山知樹, 齋藤大輔, 塩田さやか, 篠崎隆宏, 鈴木雅之, 高木信二, 俵直弘, 橋本佳, 樋口卓哉, 福田隆

    研究報告音声言語情報処理(SLP)   Vol. 2016-SLP-112 ( No. 5 )   1 - 6   2016.7

     More details

    Language:Japanese  

    researchmap

  • 声質変換における学習時の DTW 精度が性能に与える影響

    小池治憲, 能勢隆, 篠崎隆宏, 伊藤彰則

    春季研究発表会講演論文集   313 - 314   2016.3

     More details

    Language:Japanese  

    researchmap

  • 進化的戦略による高精度大語彙音声認識システムの多目的最適化

    森谷崇史, 田中智大, 篠崎隆宏, 渡部晋治, Duh Kevin

    春季研究発表会講演論文集   45 - 46   2016.3

     More details

    Language:Japanese  

    researchmap

  • 入力話者非依存ニューラルネットワークに基づく差分スペクトルフィルタを用いた声質変換における学習データ量の影響

    小池治憲, 能勢隆, 篠崎隆宏, 伊藤彰則

    春季研究発表会講演論文集   241 - 242   2016.3

     More details

    Language:Japanese  

    researchmap

  • Kaldi 用 CSJ レシピへの RNN 言語モデルの導入と性能評価

    田中智大, 森谷崇史, 篠崎隆宏, 渡部晋治, 堀貴明

    春季研究発表会講演論文集   193 - 194   2016.3

     More details

    Language:Japanese  

    researchmap

  • KaldiにおけるCSJレシピの利用法

    篠崎隆宏, 森谷崇史, 田中智大, 渡部晋治

    音声言語情報処理研究会   2016.2

     More details

    Language:Japanese  

    researchmap

  • 粒子フィルタとガウス過程回帰によるシングルチャネル音源分離

    博多屋涼, 篠崎隆宏, 郡山知樹

    研究報告音声言語情報処理(SLP)   Vol. 2016-SLP-110 ( No. 6 )   1 - 6   2016.1

     More details

    Language:Japanese  

    researchmap

  • Automation of high performance system building for large vocabulary speech recognition using evolution strategy with pareto optimality

    115 ( 346 )   31 - 36   2015.12

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • Facial image conversion based on transformation of Animation Units using DNN

    115 ( 303 )   23 - 28   2015.11

     More details

    Language:Japanese  

    researchmap

  • A Study on Speaker-Independent Voice Conversion Using Spectral Differential Filter Based on Neural Network

    115 ( 253 )   13 - 18   2015.10

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • Switch-To-Speech Communication Aid System Using WFST and Low Latency Search Algorithm

    115 ( 253 )   51 - 56   2015.10

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • 高精度日本語話し言葉音声認識のためのKaldiレシピとその評価

    森谷崇史, 篠崎隆宏, 渡部晋治

    秋季研究発表会講演論文集   155 - 156   2015.9

     More details

    Language:Japanese  

    researchmap

  • DNN特徴量抽出器に基づく単語検出器のFPGA実装と評価

    朱凱, 李昊霖, 篠崎隆宏, 堀内靖雄, 黒岩眞吾

    秋季研究発表会講演論文集   153 - 154   2015.9

     More details

    Language:Japanese  

    researchmap

  • 国際会議ICASSP2015参加報告

    岡本拓磨, 小川哲司, 落合翼, 柏木陽佑, 亀岡弘和, 木下慶介, 郡山知樹, 齋藤大輔, 篠崎隆宏, 高木信二, 滝口哲也, 太刀岡勇気, 俵直弘, 橋本佳, 藤本雅清, 松田繁樹, 三村正人, 吉岡拓也, 渡部晋治

    研究報告音声言語情報処理(SLP)   Vol. 2015-SLP-107 ( No. 3 )   1 - 7   2015.7

     More details

    Language:Japanese  

    researchmap

  • A study on speaker conversion using speech and expression features for video chatting

    115 ( 38 )   45 - 50   2015.5

     More details

    Language:Japanese  

    researchmap

  • ビデオ通話における音声および表情特徴量を用いた話者変換の検討

    齋藤優貴, 能勢 隆, 篠崎隆宏, 伊藤彰則

    EMM研究会   2015.5

     More details

    Language:Japanese  

    researchmap

  • ビデオ通話におけるニューラルネットワークを利用した話者変換の検討

    齋藤優貴, 能勢 隆, 篠崎隆宏, 伊藤彰則

    情報処理学会第77回全国大会論文集   2015.3

     More details

    Language:Japanese  

    researchmap

  • 言語モデルと音響モデルを用いた自動韻律ラベリングの評価

    増子 理菜, 郡山 知樹, 篠崎 隆宏, 小林 隆夫

    春季研究発表会講演論文集   361 - 362   2015.3

     More details

    Language:Japanese  

    researchmap

  • 進化的アルゴリズムの大規模実行によるDNN構造最適化

    篠崎 隆宏, 渡部 晋治

    春季研究発表会講演論文集   11 - 12   2015.3

     More details

    Language:Japanese  

    researchmap

  • DNN特徴量抽出器とDTWによる組み込みシステム向け耐雑音単語検出器の検討

    朱 凱, 篠崎 隆宏

    春季研究発表会講演論文集   155 - 156   2015.3

     More details

    Language:Japanese  

    researchmap

  • ニューラルネットワークを用いた話者特徴量抽出に基づく一対多クロスリンガル声質変換

    伊藤 洋二郎, 篠崎 隆宏, 能勢 隆

    春季研究発表会講演論文集   397 - 398   2015.3

     More details

    Language:Japanese  

    researchmap

  • ニューラルネットワークに基づくユーザ音声を必要としない多対一声質変換の検討

    能勢 隆, 篠崎 隆宏, 伊藤 洋二郎, 伊藤 彰則

    春季研究発表会講演論文集   271 - 274   2015.3

     More details

    Language:Japanese  

    researchmap

  • スピーキングテストシステムにおける発話内容を考慮した自動採点

    小野 豊, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    電子情報通信学会   2015.3

     More details

    Language:Japanese  

    researchmap

  • 話者特徴量入力を付加したデノイジングオートエンコーダによるクロスリンガル声質変換 (音声) -- (第16回音声言語シンポジウム)

    伊藤 洋二郎, 篠崎 隆宏, 能勢 隆

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   114 ( 365 )   13 - 18   2014.12

     More details

    Language:Japanese   Publisher:一般社団法人電子情報通信学会  

    数発話程度のごく少量のラベルなし音声を用いて特定話者の任意の発話を任意話者の声質に変換することを目的として,音声特徴量を音声特徴量に変換するデノイジングオートエンコーダに話者特徴量入力を付加した構造を持つニューラルネットを用いた声質変換手法を提案する.多言語音声コーパスを用いた実験により,提案法の有効性を示す.

    CiNii Books

    researchmap

  • 話者特徴量入力を付加したデノイジングオートエンコーダによるクロスリンガル声質変換

    伊藤洋二郎, 篠崎隆宏, 能勢隆

    音声言語情報処理研究会 (SIG-SLP)   2014.12

     More details

    Language:Japanese  

    researchmap

  • GMMに基づく声質変換のためのMDL基準による混合数の自動決定

    小林 友哉, 能勢 隆, 篠崎 隆宏, 小林 隆夫

    秋季講演論文集   341 - 342   2014.9

     More details

    Language:Japanese  

    researchmap

  • Denoising Autoencoderによる残響除去の大語彙音声認識における評価

    小宮山 大樹, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    秋季講演論文集   131 - 132   2014.9

     More details

    Language:Japanese  

    researchmap

  • ディープニューラルネットワークを用いた簡素な構造の単一単語検出器の検討

    篠崎 隆宏

    秋季講演論文集   149 - 150   2014.9

     More details

    Language:Japanese  

    researchmap

  • 眼電位入力音声合成インタフェースのためのコンテキスト依存眼動素を用いた眼電位認識

    房 福明, 篠崎 隆宏, 古井 貞煕, 堀内 靖雄, 黒岩 眞吾

    秋季講演論文集   393 - 394   2014.9

     More details

    Language:Japanese  

    researchmap

  • 複数ドメインコーパスからの文選択に基づくキャラクター音声合成の検討

    荒生 侑介, 能勢 隆, 篠崎 隆宏, 小林 隆夫

    秋季講演論文集   2014.9

     More details

    Language:Japanese  

    researchmap

  • ボルツマンマシンとMCMCサンプリングを用いた音声のシングルチャネル雑音除去

    博多屋 涼, 篠崎隆宏, 小林隆夫

    秋季研究発表会講演論文集   59 - 60   2014.9

     More details

    Language:Japanese  

    researchmap

  • スイッチ入力音声コミュニケーション支援システムのための入力プロトコル推薦手法

    房 福明, 篠崎隆宏, 小林隆夫

    秋季研究発表会講演論文集   229 - 230   2014.9

     More details

    Language:Japanese  

    researchmap

  • スイッチ入力音声合成システムのための仮名プロトコル推薦手法

    房福明, 篠崎 隆宏, 小林隆夫

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   Vol. 114 ( No. 52 )   355 - 360   2014.5

     More details

    Language:Japanese  

    researchmap

  • A Kana Protocol Recommendation Method for Switch Input Speech Synthesis Systems

    2014 ( 68 )   1 - 6   2014.5

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • ハードウエア音声認識研究のためのプラットフォームFPGA基板

    永谷 悠, 李 昊霖, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    春季講演論文集   185 - 186   2014.3

     More details

    Language:Japanese  

    researchmap

  • 腕時計型スマートデバイスにおける音声GUIの有効性の検討

    山本 宗典, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    春季講演論文集   147 - 148   2014.3

     More details

    Language:Japanese  

    researchmap

  • SCMS2.0によるタンパク質ポテンシャルエネルギー最小化の諸条件における評価

    篠崎隆宏, 関嶋政和

    バイオ情報学研究発表会   2014.3

     More details

    Language:Japanese  

    researchmap

  • 音声合成のための音韻・韻律コンテキストを考慮した文選択アルゴリズムの評価

    荒生侑介, 能勢 隆, 郡山知樹, 篠崎隆宏, 小林隆夫

    日本音響学会2014年春季研究発表会講演論文集   405 - 406   2014.3

     More details

    Language:Japanese  

    researchmap

  • HMM音声合成のための音節出現頻度にロバストな音素セットの検討

    舘野英樹, 能勢 隆, 郡山知樹, 篠崎隆宏, 小林隆夫

    日本音響学会2014年春季研究発表会講演論文集   409 - 410   2014.3

     More details

    Language:Japanese  

    researchmap

  • 音響モデルと言語モデルを利用したアクセント型・アクセント句境界の同時推定

    鈴木啓史, 郡山知樹, 能勢 隆, 篠崎隆宏, 小林隆夫

    日本音響学会2014年春季研究発表会講演論文集   441 - 442   2014.3

     More details

    Language:Japanese  

    researchmap

  • 「音声認識」は今後こうなる!

    河原達也, 篠田浩一, 堀貴明, 堀智織, 篠崎隆宏

    SIG-SLP第100回記念シンポジウム   2014.1

     More details

    Language:Japanese  

    researchmap

  • Automatic Estimation of Accent Phrase Boundaries Using Language and Acoustic Models

    Hiroshi Suzuki, Tomoki Koriyama, Takashi Nose, Takahiro Shinozaki, Takao Kobayashi

    IPSJ SIG Notes   2013 ( 16 )   1 - 6   2013.12

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    This paper proposes a technique for automatically estimating accent phrase boundaries for text-to-speech synthesis systems. To construct speech synthesis systems, we need to prepare a database that has annotations of prosodic information including accents. However, manual annotation for this purpose generally requires costly process. In contrast, the proposed method utilizes conditional random field (CRF) for the language models of accent phrase boundary and accent type, and uses hidden markov model (HMM) for the acoustic feature model. In this paper, we confirmed that the proposed method improved the estimation accuracy for reading-style speech data compared with conventional method.

    CiNii Books

    researchmap

  • 言語モデルと音響モデルを利用したアクセント境界の自動推定

    鈴木啓史, 郡山知樹, 能勢 隆, 篠崎隆宏, 小林隆夫

    電子情報通信学会技術研究報告   Vol. 113 ( No. 366 )   97 - 102   2013.12

     More details

    Language:Japanese  

    researchmap

  • S-CATにおける音響特徴量とSVRによるスコア推定

    篠崎 隆宏, 小野 豊

    日本行動計量学会   41   44 - 45   2013.9

     More details

    Language:Japanese   Publisher:日本行動計量学会  

    CiNii Books

    researchmap

  • Denoising Autoencoderを用いた残響下大語彙音声認識の検討

    小宮山 大樹, 石井 敬章, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    情報処理学会   vol. 2013-SLP-97 ( No. 1 )   1 - 6   2013.7

     More details

    Language:Japanese  

    researchmap

  • Preliminary Study of Captioning Method Considering User Characteristics

    SHIRAI Yosuke, YANAGIMURA Mai, SHINOZAKI Takahiro, HORIUCHI Yasuo, KUROIWA Shingo, ENDO Toshiki, UTSUNOMIYA Eiji

    Technical report of IEICE. Multimedia and virtual environment   vol. 112 ( no. 475 )   245 - 250   2013.3

     More details

    Language:Japanese   Publisher:一般社団法人電子情報通信学会  

    In this paper, we provide the evaluation results on the effectiveness of synchronization between voice and subtitles, and the comparative evaluation on the intelligibility and the subjective evaluation by means of the demo video with the full-text subtitle and summarized subtitle. These evaluations are conducted with the videos, which are extracted from movie films, news programs and TV travel programs. As a result, the best intelligibility has been found in the case of the voice and subtitles are perfectly synchronized. The summarized subtitles are relatively higher than the full-text subtitle in the subjective evaluation, though intelligibility in the case of the summarized subtitle is equally likely that of summarized subtitle.

    CiNii Books

    researchmap

  • Sign Language Recognition Using Kinect and Particle Filter

    FURUYA Yoshihiro, IMAMURA Daisuke, HORIUCHI Yasuo, KAWAMOTO Kazuhiko, SHINOZAKI Takahiro, KUROIWA Shingo

    Technical report of IEICE. Multimedia and virtual environment   112 ( 474 )   251 - 256   2013.3

     More details

    Language:Japanese   Publisher:一般社団法人電子情報通信学会  

    In this paper, we will discuss a sign language recognition method using a Particle Filter and Kinect. We have previously proposed an arm detection method based on a particle filter algorithm using depth and skin color information. We have implemented the method using Kinect and demonstrated that it gave good recognition accuracy. However, the method has a constraint that the users have to roll up their sleeves since it requires the color of arms. In this study, we propose an improved algorithm that removes the constraint. Experimental results show that the new algorithm gives comparable performance as the previous one without using the arm color.

    researchmap

  • Eye Motion Input Based Speech Synthesis Interface for Communication Aids

    FANG Fuming, SHINOZAKI Takahiro, HORIUCHI Yasuo, KUROIWA Shingo, FURUI Sadaoki, MUSHA Toshimitsu

    IEICE technical report. Welfare Information technology   112 ( 426 )   29 - 34   2013.2

     More details

    Language:Japanese   Publisher:一般社団法人電子情報通信学会  

    In order to provide an efficient means of communication for those who cannot move muscles of their whole body except eyes due to amyotrophic lateral sclerosis (ALS), we are studying a speech synthesis interface based on electrooculogram (EOG) input The system consists of an EOG input module, an eye motion recognizer, and a speech synthesizer In this paper, we improve the EOG input based eye motion recognizer applying speech recognition techniques In our previous system, a hidden Markov model (HMM) based bi eye-motion model was used However, it was not enough to effectively model the context effects of eye motions In this study, we investigate using a tied-state tri eye-motion model Moreover, an N-gram model is integrated to the recognition system In the experiment, it is shown that 96 2% of character recognition accuracy is obtained by using the tn eye-motion model whereas it is 84 3% and 89 1% for mono and bi eye-motion models, respectively By using a character 3-gram model in combination with the tri eye motion-model, the highest character accuracy of 97 3% has been obtained

    CiNii Books

    researchmap

  • 音声認識システムのパイプライン分解と遅延評価を用いた実装法

    篠崎隆宏, 古井貞熙, 堀内靖雄, 黒岩眞吾

    日本音響学会2012年秋季研究発表会   2012.9

     More details

    Language:Japanese  

    researchmap

  • 日本語スピーキングテストにおける文章読み上げ問題の自動採点の検討

    山畑 勇人, 大久保 梨思子, 山田 武志, 今井 新悟, 石塚 賢吉, 篠崎 隆宏, 西村 竜一, 牧野 昭二, 北脇 信彦

    秋季講演論文集   399 - 400   2012.9

     More details

    Language:Japanese  

    researchmap

  • コミュニケーション支援のための連続眼電位認識の研究

    房福明, 篠崎隆宏, 古井貞熙, 堀内靖雄, 黒岩眞吾

    日本音響学会2012年秋季研究発表会   1513 - 514   2012.9

     More details

    Language:Japanese  

    researchmap

  • 日本語スピーキングテストシステムS-CAT のためのSVR による自由発話の自動採点

    小野 豊, 大竹 美鈴, 篠崎 隆宏, 西村 竜一, 山田 武志, 石塚 賢吉, 堀内 靖雄, 黒岩 眞吾, 今井 新悟

    秋季講演論文集   335 - 336   2012.9

     More details

    Language:Japanese  

    researchmap

  • 日本語スピーキングテストにおける文生成問題の自動採点の検討

    大久保 梨思子, 山畑 勇人, 山田 武志, 今井 新悟, 石塚 賢吉, 篠崎 隆宏, 西村 竜一, 牧野 昭二, 北脇 信彦

    秋季講演論文集   395 - 396   2012.9

     More details

    Language:Japanese  

    researchmap

  • 純粋関数型コンパクトデコーダHusky2 の性能評価

    深津 澪, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    秋季講演論文集   187 - 188   2012.9

     More details

    Language:Japanese  

    researchmap

  • 日本語スピーキングテストS-CAT における並列セグメンテーションを用いた自動採点の検討

    西村 竜一, 栗原 理沙, 篠崎 隆宏, 石塚 賢吉, 山田 武志, 今井 新悟, 河原 英紀, 入野 俊夫

    秋季講演論文集   397 - 399   2012.9

     More details

    Language:Japanese  

    researchmap

  • New Speech Research Paradigm in the Cloud Era

    Tomoyoshi Akiba, Koji Iwano, Jun Ogata, Tetsuji Ogawa, Nobutaka Ono, Takahiro Shinozaki, Koichi Shinoda, Hiroaki Nanjo, Hiromitsu Nishizaki, Masafumi Nishida, Ryuichi Nishimura, Sunao Hara, Takaaki Hori

    IPSJ SIG Notes   Vol. 2012-SLP-92 ( No. 4 )   1 - 7   2012.7

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    Recently most individuals have come to use mobile information devices, and daily upload the information obtained by such devices to Internet Cloud. Accordingly the applications of speech information processing have been changing drastically. We need to create a new paradigm for the research and development of speech information processing to adapt to this change. In this paper, we summarize the state-of-the-art speech technologies, propose how to create a research platform for this new paradigm, and discuss the problems we should solve to realize it.

    CiNii Books

    researchmap

  • Slice Chain Max-Sumアルゴリズムによるタンパク質のポテンシャルエネルギー最小化に関する研究

    猪瀬直人, 篠崎隆宏, 杜世橋, 古井貞熙, 関嶋政和

    情報処理学会バイオ情報学研究会   Vol. 2012-BIO-28 ( No. 20 )   1 - 8   2012.3

     More details

    Language:Japanese  

    researchmap

  • 日本語スピーキングテストにおける文章読み上げ問題の採点に影響を及ぼす要因の検討

    山畑 勇人, 大久保 梨思子, 山田 武志, 今井 新悟, 石塚 賢吉, 篠崎 隆宏, 西村 竜一, 牧野 昭二, 北脇 信彦

    電子情報通信学会総合大会   2012.3

     More details

    Language:Japanese  

    researchmap

  • 眼電位入力音声合成インタフェースの提案とユーザー適応の検討

    房福明, 篠崎隆宏, 堀内靖雄, 黒岩眞吾, 古井貞熙, 武者利光

    第39回知能システムシンポジウム資料   293 - 298   2012.3

     More details

    Language:Japanese  

    researchmap

  • 言語モデルの順向き最尤文選択適応への教師なしクロスバリデーション適応法の応用

    篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    春季講演論文集   99 - 100   2012.3

     More details

    Language:Japanese  

    researchmap

  • AWA長期間収録音声コーパスと時期差の分析

    黒岩 眞吾, 柘植 覚, 張 文彬, 篠崎 隆宏, 堀内 靖雄

    春季講演論文集   83 - 86   2012.3

     More details

    Language:Japanese  

    researchmap

  • ストーリー性を考慮した映画あらすじからの類似度計算

    村手宏輔, 黒岩眞吾, 堀内靖雄, 篠崎隆宏

    全国大会講演論文集   2012 ( 1 )   535 - 537   2012.3

     More details

    Language:Japanese   Publisher:一般社団法人情報処理学会  

    情報推薦に用いられるコンテンツベースベース技術に関して、あらすじが書かれた文書などストーリー性のあるコンテンツに対する類似度計算方法を提案する.ストーリーとは映画や小説などに含まれる話の筋のことであり、それらを説明する文書の中では人物の行動の経緯など要素の連続によって表現されていることが多い.しかし、従来の文書間類似度を計算する際に用いられるベクトル空間モデルでは、出現順序によって意味合いが変るストーリーを比較することは難しい.本研究ではストーリー性を考慮した文書の比較を行うことを目標とし、映画のあらすじ文書を対象に要素の並びを利用した類似度計算方法を検討した.

    CiNii Books

    researchmap

  • Multimodal Speech Recognition Based on Lightweight Visual Features Reviewed

    YOSHIKAWA Masayoshi, SHINOZAKI Takahiro, IWANO Koji, FURUI Sadaoki

    The IEICE transactions on information and systems (Japanese edetion)   Vol. J95-D ( No. 3 )   618 - 627   2012.3

     More details

    Language:Japanese   Publisher:The Institute of Electronics, Information and Communication Engineers  

    CiNii Books

    researchmap

  • HMM Sign Language Recognition Using Kinect and Particle Filter

    NISHIMURA Yosuke, IMAMURA Daisuke, HORIUCHI Yasuo, KAWAMOTO Kazuhiko, SHINOZAKI Takahiro, KUROIWA Shingo

    IEICE technical report. Speech   vol. 111 ( no. 431 )   161 - 166   2012.2

     More details

    Language:Japanese   Publisher:一般社団法人電子情報通信学会  

    In this paper, we will introduce a sign language recognition method using Kinect which is a motion sensing input device by Microsoft. Kinect has a RGB camera and a depth sensor and therefore we can easily get 3D images of signers' motion. The positions of arms are detected from both the color image data and the depth data of each pixel using Particle Filter Algorithm. Then, sign sentences are recognized using Hidden Markov Model. In our method, the recognition rate was 86.0%, while the recognition rate in the previous study using video image only was 76.2%. On the other hand, the recognition rate using attached motion capturing sensors was 86.8% and was approximately the same as our method. These results show that our method is useful for the practical applications, since our method uses only Kinect which is not expensive and no device is attached to the signer's hand.

    J-GLOBAL

    researchmap

  • 日本語発話能力測定ウェブシステムのための留学生発話分析

    栗原 理沙, 石塚 賢吉, 西村 竜一, 篠崎 隆宏, 山田 武志, 今井 新悟

    信学技報   vol. 111 ( no. 431 )   141 - 142   2012.2

     More details

    Language:Japanese  

    researchmap

  • Electrooculogram recognition using hidden Markov model

    FANG Fuming, SHINOZAKI Takahiro, HORIUCHI Yasuo, KUROIWA Shingo, FURUI Sadaoki, MUSHA Toshimitsu

    IEICE technical report. Speech   111 ( No. SP2011-117 )   97 - 102   2012.2

     More details

    Language:Japanese   Publisher:一般社団法人電子情報通信学会  

    In order to provide an efficient means of communication for those who cannot move muscles of their whole body except eyes due to amyotrophic lateral sclerosis (ALS), we propose an speech synthesis interface based on electrooculogram (EOG) input. The system consists of EOG electrodes, an EOG recognition system, and a speech synthesis system. In this paper, we report experiments about the EOG recognition system that we have developed borrowing speech recognition techniques using hidden Markov model (HMM). In the experiments, we first make user-dependent EOG recognition systems. It is shown that the systems give 95.7% recognition accuracy on average. While they give high recognition performance, a problem is that they need a large amount of user-specific data for model training. From the application point of view, user-independent systems are preferable. As the second experiment, we evaluate the effect of individual differences in EOG recognition. It is shown that the recognition accuracy largely drops if there is a mismatch between the EOG model and recognition data. As the last experiment, we apply speaker adaptation techniques that have been developed for speech recognition to EOG recognition, and show that they are effective to improve EOG recognition accuracy.

    CiNii Books

    researchmap

  • Comparative Analysis of Turn-taking between Japanese Sign Language and Japanese Speech

    MURASE Yumi, HORIUCHI Yasuo, SHINOZAKI Takahiro, KUROIWA Shingo

    IEICE technical report. Welfare Information technology   111 ( 424 )   7 - 12   2012.1

     More details

    Language:Japanese   Publisher:The Institute of Electronics, Information and Communication Engineers  

    In this research, we analyzed turn-taking phenomena in spontaneous dialogue comparing Japanese Sign Language (JSL) and Japanese oral language (JOL) based on the turn-taking rules for oral language by Sacks et al. Three dialogue data by six native signers of JSL and three dialogue data by six native speakers of JOL were used for the analysis (each dialogue was about 5 minutes long). As a result, it was suggested that JSL followed the turn-taking rules similarly to JOL, while overlap duration in JSL was longer than in JOL. Two reasons were found: (1) when overlap occurred in JOL, the original speaker had a tendency to stop his/her utterance, but in JSL, he/she continued his/her utterance to the end, (2) signers in JSL sometimes repeated or restated his/her utterance after TRP and this resulted in overlap with next signers. However, in the situation (2), it was observed that the signer released his/her turn after TRP by lacking or weakening NMSs (non manual signals). From these results, we discussed the influence caused by differences between visual language and oral language.

    CiNii Books

    researchmap

  • Protein Potential Energy Minimization Using Slice Chain Max-Sum Algorithm

    N. Inose, T. Shinozaki, S. Du, S. Furui, M. Sekijima

    26th Annual Symposium of The Protein Society   2012

     More details

    Language:English  

    researchmap

  • Distance-based factor graph linearization and sampled max-sum algorithm for efficient 3D potential decoding of macromolecules Reviewed

    Takahiro Shinozaki, Toshinao Iwaki, Shiqiao Du, Masakazu Sekijima, Sadaoki Furui

    IPSJ Transaction on Bioinformatics   Vol. 4 ( 1 )   34 - 44   2011.12

     More details

    Language:English   Publisher:Information and Media Technologies Editorial Board  

    Three-dimensional structure prediction of a molecule can be modeled as a minimum energy search problem in a potential landscape. Popular ab initio structure prediction approaches based on this formalization are the Monte Carlo methods represented by the Metropolis method. However, their prediction performance degrades for larger molecules such as proteins since the search space is exponential to the number of atoms. In order to search the exponential space more efficiently, we propose a new method modeling the potential landscape as a factor graph. The key ideas are slicing the factor graph based on the maximum distance of bonded atoms to convert it to a linear structured graph, and the utilization of the max-sum search algorithm combined with samplings. It is referred to as Slice Chain Max-Sum and it has an advantage that the search is efficient because the graph is linear. Experiments are performed using polypeptides having 50 to 300 amino acid residues. It has been shown that the proposed method is computationally more efficient than the Metropolis method for large molecules.

    DOI: 10.2197/ipsjtbio.4.34

    researchmap

  • 時期差に頑健な話者識別手法

    張 文彬, 陸 昊澤, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    バイオメトリクスと認識・認証シンポジウム   2011.11

     More details

    Language:Japanese  

    researchmap

  • 構内アナウンス環境下における音声認識のための音声区間検出

    紺野 遼輔, 篠崎 隆宏, 堀内 靖雄, 黒岩 眞吾

    日本音響学会   151 - 152   2011.9

     More details

    Language:Japanese  

    researchmap

  • Distance-based Graph Linearization and Sampled Max-sum Algorithm for Efficient 3D Potential Decoding of Macromolecules

    2011 ( 5 )   1 - 8   2011.9

     More details

    Language:English  

    CiNii Books

    researchmap

  • Sampled Max-Sum Algorithm and Application to 3D Structure Prediction of Proteins

    岩木 聡直, 篠崎 隆宏, 古井貞熙

    日本蛋白質科学会年会   2011.6

     More details

    Language:Japanese  

    researchmap

  • 純粋関数型言語を用いた超コンパクトデコーダの開発

    篠崎隆宏, 関嶋政和, 萩原茂樹, 古井貞熙

    情報処理学会   2011.4

     More details

    Language:Japanese  

    researchmap

  • N-gramカウントを用いた言語モデルの効率的な選択学習

    久保田 雄, 篠崎 隆宏, 古井 貞熙, 宇都宮 栄二, 新堂 安孝

    日本音響学会2011年春季講演論文集   ( No. 3-5-2 )   73 - 74   2011.3

     More details

    Language:Japanese  

    researchmap

  • クロス言語検索を用いた中国語音声認識による乗換案内システム

    張 ?, 大西 翼, 篠崎 隆宏, 古井 貞熙

    日本音響学会2011年春季講演論文集   ( No. 2-5-7 )   61 - 62   2011.3

     More details

    Language:Japanese  

    researchmap

  • 眼電位を用いた音声合成インタフェースの研究

    尾崎 賢人, 篠崎 隆宏, 武者 利光, 古井 貞煕

    日本音響学会2011年春季講演論文集   ( No. 3-4-13 )   1621 - 1622   2011.3

     More details

    Language:Japanese  

    researchmap

  • ホームビデオからのハイライト検出支援のための音声情報の視覚化

    高木 幸一, 川田 亮一, 篠崎 隆宏, 古井 貞熙

    日本音響学会2010年秋季講演論文集   ( No. 2-9-11 )   69 - 70   2010.9

     More details

    Language:Japanese  

    researchmap

  • 柔軟でコンパクトな純粋関数型デコーダの検討

    篠崎 隆宏, 関嶋 政和, 萩原 茂樹, 古井 貞熙

    日本音響学会2010年秋季講演論文集   ( No. 1-Q-26 )   181 - 182   2010.9

     More details

    Language:Japanese  

    researchmap

  • Home video trimming method based on a difference depending on presence or absence of audio signals

    IEICE technical report   110 ( 128 )   51 - 56   2010.7

     More details

    Language:Japanese  

    researchmap

  • Home Video Trimming Method based on a Difference Depending on Presence or Absence of Audio Signals

    TAKAGI Koichi, KAWADA Ryoichi, SHINOZAKI Takahiro, FURUI Sadaoki

    2010 ( 10 )   1 - 6   2010.7

     More details

  • 年齢推定のための音声特徴量および推定器の検討

    和田 俊也, 篠崎 隆宏, 古井 貞熙

    電子情報通信学会 技術研究報告   Vol. SP2010-27   31 - 36   2010.6

     More details

    Language:Japanese  

    researchmap

  • 識別学習モデルと教師なしCV適応を用いたCSJ講演音声認識

    篠崎 隆宏, 久保田 雄, ディクソン・ポール, 古井 貞煕

    日本音響学会2010年春季講演論文集   ( No. 1-6-14 )   37 - 38   2010.3

     More details

    Language:Japanese  

    researchmap

  • MLLR変換行列を特徴量として用いた年齢推定

    和田俊也, 篠崎隆宏, 古井貞熙

    日本音響学会2010年春季講演論文集   ( No. 2-6-13 )   83 - 84   2010.3

     More details

    Language:Japanese  

    researchmap

  • 自然性と個人性に優れた音声合成のための音素継続時間長適応法

    神山歩相名, 篠崎隆宏, 岩野公司, 古井貞熙

    日本音響学会2010年春季講演論文集   ( No. 2-7-1 )   329 - 330   2010.3

     More details

    Language:Japanese  

    researchmap

  • 日本語話し言葉コーパスを用いた異なるタスクに対する音声認識

    西井 俊介, 篠崎 隆宏, 古井 貞熙

    日本音響学会2010年春季講演論文集   ( No. 1-6-10 )   27 - 28   2010.3

     More details

    Language:Japanese  

    researchmap

  • User identification using Time-of-Flight camera image streams

    Felipe Gomez-Caballero, Takahiro Shinozaki, Sadaoki Furui

    ( No. 5X-8 )   2 - 615   2010.3

     More details

    Language:English  

    researchmap

  • HMM音声合成における自然性と個人性に優れた韻律モデル適応法の検討

    神山 歩相名, 篠崎 隆宏, 岩野 公司, 古井 貞煕

    情報処理学会研究会報告   Vol. 2010-SLP-80 ( No. 12 )   1 - 6   2010.2

     More details

    Language:Japanese  

    researchmap

  • 教師無しアンサンブル適応法の提案と音響モデル適応への応用

    篠崎 隆宏, 古井 貞煕

    第12回情報論的学習理論ワークショップ   2009.10

     More details

    Language:Japanese  

    researchmap

  • 目的音GMM尤度基準スペクトル補正法の諸評価

    篠崎 隆宏, 古井 貞熙

    日本音響学会2009年秋季講演論文集   ( No. 1-1-10 )   31 - 32   2009.9

     More details

    Language:Japanese  

    researchmap

  • 自然性と個人性に優れたF0パターン適応法

    神山 歩相名, 篠崎 隆宏, 岩野 公司, 古井 貞熙

    日本音響学会2009年秋季講演論文集   ( No. 1-2-7 )   249 - 250   2009.9

     More details

    Language:Japanese  

    researchmap

  • 音響モデルのアンサンブル学習

    篠崎 隆宏

    ( No. 11. )   2009.7

     More details

    Language:Japanese  

    researchmap

  • 教師なしクロスバリデーション適応法の諸条件における評価

    久保田 雄, 篠崎 隆宏, 古井 貞熙

    "情報処理学会研究報告, IPSJ SIG Technical Report"   Vol. 2009-SLP-77 ( No. 7 )   2009.7

     More details

    Language:Japanese  

    researchmap

  • F0パターン生成モデルのための数量化?類の平均値置換による話者適応法の検討

    神山 歩相名, 篠崎 隆宏, 岩野 公司, 古井 貞熙

    電子情報通信学会 技術研究報告   87 - 92   2009.6

     More details

    Language:Japanese  

    researchmap

  • 高精度音声認識のための教師なしクロスバリデーション適応法の提案

    篠崎 隆宏, 久保田 雄, 古井貞熙

    日本音響学会2009年春季講演論文集   ( No. 1-5-10 )   27 - 28   2009.3

     More details

    Language:Japanese  

    researchmap

  • 教師なしクロスバリデーション適応によるタスク適応

    久保田 雄, 篠崎 隆宏, 古井貞熙

    日本音響学会2009年春季講演論文集   ( No. 1-5-11 )   29 - 30   2009.3

     More details

    Language:Japanese  

    researchmap

  • 音声による3次元直接操作インタフェース Reviewed

    川崎智久, 大西 翼, 篠崎 隆宏, 古井貞熙

    インタラクション2009   43 - 44   2009.3

     More details

    Language:Japanese  

    researchmap

  • 高精度音声認識のための教師なしクロスバリデーションおよび集合適応法の提案

    篠崎 隆宏, 久保田 雄, 古井貞熙

    社団法人 情報処理学会 研究報告 (2009-SLP-75)   ( No. 75 )   1 - 6   2009.2

     More details

    Language:Japanese  

    researchmap

  • 携帯端末上でのプロキシ編集

    高木 幸一, 米山 暁夫, 篠崎 隆宏, 古井貞熙

    電子情報通信学会 技術研究報告   ( No. IE2009-02 )   7 - 12   2009.2

     More details

    Language:Japanese  

    researchmap

  • 音声入力によるマウスの直接操作の検討

    川崎 智久, 大西 翼, 岩野 公司, 篠崎 隆宏, 古井貞熙

    日本音響学会2008年秋季講演論文集   ( No. 1-1-23 )   55 - 56   2008.9

     More details

    Language:Japanese  

    researchmap

  • 目的音GMMを用いたスペクトル補正フィルタの提案

    篠? 隆宏, 古井 貞煕

    日本音響学会2008年秋季講演論文集   ( No. 1-1-1 )   1 - 2   2008.9

     More details

    Language:Japanese  

    researchmap

  • 効率的なクロスバリデーションに基づく混合ガウス分布の最適化とその拡張

    篠? 隆宏, 古井 貞煕, 河原 達也

    社団法人 情報処理学会 研究報告   2008-SLP-72   69 - 74   2008.7

     More details

    Language:Japanese  

    researchmap

  • クロスバリデーション尤度によるHMMの混合数の最適化

    篠崎 隆宏, 河原 達也

    春季講演論文集   41 - 42   2008.3

     More details

    Language:Japanese  

    researchmap

  • Aggregated cross-validation尤度を用いた混合ガウス分布最適化アルゴリズムの提案

    篠崎 隆宏, 古井 貞熙, 河原 達也

    日本音響学会2008年春季講演論文集   ( No. 2-10-1 )   67 - 68   2008.3

     More details

    Language:Japanese  

    researchmap

  • Initial Evaluation of the Drivers' Japanese Speech Corpus in a Car Environment

    Kousuke Hiraki, Takahiro Shinozaki, Koji Iwano, Agnieszka Betkowska, Betkowska Agnieszka, Koichi Shinoda, SADAOKI FURUI

    Vol. SP2007-202   93 - 98   2008.3

     More details

    Language:English  

    researchmap

  • 頑健なパラメタ推定のためのAggregated EM 法の提案と評価

    篠崎 隆宏, Mari Ostendorf, 河原 達也

    電子情報通信学会 技術研究報告   223 - 228   2007.12

     More details

    Language:Japanese  

    researchmap

  • 頑健なパラメタ推定のためのAggregated EMアルゴリズムの提案

    篠崎 隆宏, Mari Ostendorf, 河原 達也

    秋季講演論文集   131 - 134   2007.9

     More details

    Language:Japanese  

    researchmap

  • 効率的なクロスバリデーション尤度評価に基づく混合ガウス分布の最適化

    篠崎 隆宏, 河原 達也

    情報処理学会   81 - 86   2007.7

     More details

    Language:Japanese  

    researchmap

  • ICASSP2007報告

    戸田 智基, 篠崎 隆宏, 秋田 祐哉

    情報処理学会   45 - 48   2007.7

     More details

    Language:Japanese  

    researchmap

  • 超並列計算機を用いた話し言葉音声認識の研究

    篠崎 隆宏, 河原 達也

    京都大学学術情報メディアセンター全国共同利用版[公報]   Vol. 6 ( No. 1 )   31 - 37   2007.3

     More details

    Language:Japanese  

    researchmap

  • Cross-validation EM Algorithm for Robust Parameter Estimation

    SHINOZAKI Takahiro, OSTENDORF Mari

    IPSJ SIG Notes   2006 ( 136 )   191 - 196   2006.12

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    A new maximum likelihood training algorithm is proposed that compensates for weaknesses of the EM algorithm by using cross-validation likelihood in the expectation step to avoid overtraining. By usitlg a set of sufficient statistics associated with a partitioning of the training data, as in parallel EM, the algorithm has the same order of computational requirements as the original EM algorithm. Analyses using a GMM with artificial data show the proposed algorithm is more robust for overtraining than the conventional EM algorithm. Large vocabulary recognition experiments on Mandarin broadcast news data show that the method makes better use of more parameters and gives lower recognition error rates than EM training.

    CiNii Books

    researchmap

    Other Link: http://id.nii.ac.jp/1001/00056862/

  • 頑健なパラメタ推定のためのクロスバリデーションEM法の提案

    篠崎 隆宏, Mari Ostendorf

    電子情報通信学会 技術研究報告   13 - 18   2006.12

     More details

    Language:Japanese  

    researchmap

  • State-of-the-art Technology of Speech Information Processing:Statistical Approach for Acoustic Modeling and Its Application to Speech Recognition

    SHINODA Koichi, SHINOZAKI Takahiro

    IPSJ Magazine   45 ( 10 )   1012 - 1019   2004.10

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    CiNii Books

    researchmap

    Other Link: http://id.nii.ac.jp/1001/00065158/

  • Dynamic Bayesian Network-Based Acoustic Models Incorporating Speaking Rate Effects

    SHINOZAKI Takahiro, FURUI Sadaoki

    IEICE Trans. Inf. & Syst.   87 ( 10 )   2339 - 2347   2004.10

     More details

    Language:English   Publisher:The Institute of Electronics, Information and Communication Engineers  

    One of the most important issues in spontaneous speech recognition is how to cope with the degradation of recognition accuracy due to speaking rate fluctuation within an utterance. This paper proposes an acoustic model for adjusting mixture weights and transition probabilities of the HMM for each frame according to the local speaking rate. The proposed model is implemented along with variants and conventional models using the Bayesian network framework. The proposed model has a hidden variable representing variation of the "mode" of the speaking rate, and its value controls the parameters of the underlying HMM. Model training and maximum probability assignment of the variables are conducted using the EM/GEM and inference algorithms for the Bayesian networks. Utterances from meetings and lectures are used for evaluation where the Bayesian network-based acoustic models are used to rescore the likelihood of the N-best lists. In the experiments, the proposed model indicated consistently higher performance than conventional HMMs and regression HMMs using the same speaking rate information.

    CiNii Books

    researchmap

  • 周波数帯域ごとの重みつき尤度を用いた音声認識の検討

    西村 義隆, 篠崎 隆宏, 岩野 公司, 古井 貞煕

    日本音響学会 2004年春季講演論文集   1 ( No. 2-11-9 )   117 - 118   2004.3

     More details

    Language:Japanese   Publisher:日本音響学会  

    researchmap

  • 超並列デコーダを用いた話し言葉音声認識

    篠崎 隆宏, 古井 貞熙

    日本音響学会 2004年春季講演論文集   ( No. 2-11-6 )   111 - 112   2004.3

     More details

    Language:Japanese  

    researchmap

  • 超並列デコーダによる話し言葉音声認識

    篠崎 隆宏, 古井 貞熙

    第3回話し言葉の科学と工学ワークショップ 講演予稿集   67 - 72   2004.2

     More details

    Language:Japanese  

    researchmap

  • 話し言葉音声認識へのベイジアンネットの適用

    篠崎 隆宏, 古井 貞熙

    国立国語研究所公開研究発表会 「話し言葉のデータベース ?『日本語話し言葉コーパス』?」 講演予稿集   47 - 48   2003.12

     More details

    Language:Japanese  

    researchmap

  • 周波数帯域ごとの重みつき尤度を用いた雑音に頑健な音声認識

    西村 義隆, 篠崎 隆宏, 岩野 公司, 古井 貞熙

    電子情報通信学会 技術研究報告   ( No. SP2003-116 )   19 - 24   2003.12

     More details

    Language:Japanese  

    researchmap

  • 隠れモードベイズ分類器を用いた音響モデルの適応学習

    篠崎 隆宏, 古井 貞熙

    日本音響学会 2003年秋季講演論文集   ( No. 2-6-2 )   63 - 64   2003.9

     More details

    Language:Japanese  

    researchmap

  • 重みつきスペクトル特徴量を用いた雑音に頑健な音声認識

    西村 義隆, 篠崎 隆宏, 岩野 公司, 古井 貞熙

    日本音響学会 2003年秋季講演論文集   ( No. 1-6-3 )   5 - 6   2003.9

     More details

    Language:Japanese  

    researchmap

  • Hidden Mode HMM for Speaking Rate Variation : Application of Bayesian Networks for Speech Recognition

    33 ( 4 )   245 - 250   2003.6

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • 発話速度変動を考慮した隠れモードHMMによる音声のモデル化

    篠崎隆宏, 古井 貞熙

    電子情報通信学会 技術研究報告   ( No. SP2003-41 )   37 - 42   2003.6

     More details

    Language:Japanese  

    researchmap

  • 大語彙連続音声認識のための言語的音響的属性に基づく単語単位の最適化

    篠崎隆宏, 古井貞熙

    日本音響学会 2003年春季講演論文集   ( No. 3-4-4 )   135 - 136   2003.3

     More details

    Language:Japanese  

    researchmap

  • 言語モデルの教師なしバッチ型話題適応

    横山忠介, 篠崎隆宏, 岩野公司, 古井 貞熙

    日本音響学会 2003年春季講演論文集   ( No. 3-4-1 )   129 - 130   2003.3

     More details

    Language:Japanese  

    researchmap

  • 隠れモードHMMによる発話速度変動を考慮した音声のモデル化

    篠崎 隆宏, 古井 貞熙

    日本音響学会 2003年秋季講演論文集   ( No. 2-6-1 )   61 - 62   2003

     More details

    Language:Japanese  

    researchmap

  • Unsupervised batch - type adaptation method for language models

    YOKOYAMA Tadasuke, SHINOZAKI Takahiro, IWANO Koji, FURUI Sadaoki

    IPSJ SIG Notes   2002 ( 121 )   183 - 188   2002.12

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    This paper proposes an unsupervised, batch-type, class-based language model adaptation method for spontaneous speech recognition. The word classes are automatically determined by maximizing the bigram likelihood using a training set. A class-based language model is built based on recognition hypotheses obtained using a general word-based language model, and linearly interpolated with the general language model. All the input utterances are re-recognized using the adapted language model. The proposed method was applied to the recognition of spontaneous presentations and was found to be effective in improving the recognition accuracy for all the presentations. The best condition was found to be using 100 word classes, and in this condition 2.3% of the absolute value improvement in the word accuracy averaged over all the speakers was achieved, using speaker independent acoustic models. It was also found that effectiveness of the proposed method is additive to that of the acoustic model adaptation. Consequently, 71.8% word recognition accuracy was achieved for spontaneous presentations after adapting both acoustic and language models.

    CiNii Books

    researchmap

    Other Link: http://id.nii.ac.jp/1001/00057297/

  • 言語モデルのバッチ型教師なし適応化法

    横山忠介, 篠崎隆宏, 岩野公司, 古井貞熙

    電子情報通信学会 技術研究報告   Vol. NLC2002-74 ( No. SP2002-151 )   19 - 24   2002.12

     More details

    Language:Japanese  

    researchmap

  • 講演音声認識を対象とした言語モデルの話者適応化

    横山 忠介, 篠崎 隆宏, 古井 貞熙

    日本音響学会 2002年秋季講演論文集   ( No. 3-9-6 )   141 - 142   2002.9

     More details

    Language:Japanese  

    researchmap

  • 話し言葉音声中の単語認識における人を基準としたデコーダの性能評価

    篠崎 隆宏, 古井 貞熙

    日本音響学会 2002年秋季講演論文集   ( No. 2-9-13 )   87 - 88   2002.9

     More details

    Language:Japanese  

    researchmap

  • 話し言葉音声認識における認識率の変動要因の分析と認識単位の設計

    篠崎 隆宏, 古井 貞熙

    第2回 話し言葉の科学と工学ワークショップ講演予稿集   59 - 64   2002.3

     More details

    Language:Japanese  

    researchmap

  • 話し言葉音声認識における認識性能の個人差の解析

    篠崎 隆宏, 古井 貞熙

    日本音響学会 2002年春季講演論文集   ( No. 1-5-9 )   17 - 18   2002.3

     More details

    Language:Japanese  

    researchmap

  • Presentation Transcription Using a Japanese Spontaneous Speech Corpus

    Takahiro Shinozaki, Sadaoki Furui

    43 ( 7 )   2098 - 2107   2002

     More details

  • A statistical analysis of individual differences in spontaneous speech recognition performance

    SHINOZAKI Takahiro, FURUI Sadaoki

    IPSJ SIG Notes   2001 ( 123 )   111 - 116   2001.12

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    This paper reports results of various investigations on recognizing spontaneous presentation speech. Individual differences in the speech recognition performances are analyzed. A restricted set of the speaker attributes comprising the speaking rate, the out of vocabulary rate and the repair rate is found to be most significant to yield individual differences in the word accuracy. It is shown that unsupervised MLLR speaker adaptation works well for improving the word accuracy but does not compensate for the effect of the speaking rate.

    CiNii Books

    researchmap

    Other Link: http://id.nii.ac.jp/1001/00057386/

  • 話し言葉音声認識における話者間の認識率変動要因の解析

    篠崎 隆宏, 古井 貞熙

    電子情報通信学会 技術研究報告   Vol. SP2001-102 ( No. NLC2001-67 )   1 - 6   2001.12

     More details

    Language:Japanese  

    researchmap

  • Recognition error analysis of spontaneous speech using decision trees.

    SHINOZAKI Takahiro, FURUI Sadaoki

    2001 ( No. 1-1-9 )   17 - 18   2001.10

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • Automatic speech recognition using a spontaneous speech coupus.

    SHINOZAKI Takahiro, HOSOKAWA Takao, FURUI Sadaoki

    2001 ( No. 1-3-14 )   31 - 32   2001.3

     More details

    Language:Japanese  

    CiNii Books

    researchmap

  • 話し言葉音声認識のための音響・言語モデル

    篠崎隆宏, 堀智織, 古井貞熙

    話し言葉の科学と工学ワークショップ予稿集   101 - 108   2001.3

     More details

    Language:Japanese  

    researchmap

  • Toward Spontaneous Speech Recognition

    SHINOZAKI Takahiro, SAITO Yohei, HORI Chiori, FURUI Sadaoki

    IPSJ SIG Notes   2000 ( 119 )   125 - 130   2000.12

     More details

    Language:Japanese   Publisher:Information Processing Society of Japan (IPSJ)  

    This paper reports various investigations on recognizing spontaneous speech such as lectures, interviews and discussions conducted in relation with our national project started in 1999. Usefulness of acoustic and linguistic modeling based on actual spontaneous speech corpora, registration of new words using past broadcast news or a textbook related to the areas of topics, and an acoustic backing-off method for the periods of cross talk in interviews have been confirmed. Recognition accuracy has a wide speaker-to-speaker variability according to the speaking rate, number of fillers, number of repairs, etc. This paper also reports a method for efficiently making minutes of meetings based on interaction between a speech recognition system and a user. The recognition accuracy for spontaneous speech is still very low, and there exist a large number of research issues including how to extract pseudo-sentence unit speech for recognition, how to build pronunciation dictionaries, and how to transcribe spontaneous speech in corpora.

    CiNii Books

    researchmap

    Other Link: http://id.nii.ac.jp/1001/00057471/

  • 話し言葉音声の認識を目指して

    篠崎隆宏, 斎藤洋平, 堀智織, 古井貞熙

    電子情報通信学会 技術研究報告   ( No. SP2000-96 )   7 - 12   2000.12

     More details

    Language:Japanese  

    researchmap

  • k-制限最小値独立置換族のサイズ均等性

    篠崎 隆宏, 武井 由智, 伊東 利哉

    平成12年度信越支部大会   2000.10

     More details

    Language:Japanese  

    researchmap

  • An Optimal Construction of Exactly Min-Wise Independent Permutations

    TAKEI YOSHINORI, ITOH TOSHIYA, SHINOZAKI TAKAHIRO

    IEICE technical report. Theoretical foundations of Computing   98 ( 432 )   89 - 98   1999.11

     More details

    Language:English   Publisher:The Institute of Electronics, Information and Communication Engineers  

    A family of min-wise independent permutations C is known to be a useful tool of indexing replicated documents on the Web. For any integer n>0, a family of permutations C on{1, 2, ..., n}is said to be min-wise independent if for any(nonempty)X⊆{1, 2, ..., n}and any x∈X, Pr(min{π(X)}=π(x))=∥X∥^<-1>when π is chosen uniformly at random from C, where ∥A∥is the cardinality of a finite set A. For any integer n>0, it has been known that∥c∥>1cm(n, n-1, ..., 2, 1)=e^<n-o(n)>for any family of min-wise independent permutations C on{1, 2, ..., n}and that there exists a family of min-wise independent permutations C on{1, 2, ..., n}such that∥C∥<4^n. However, it has been unclear whether there exists a family of min-wise independent family C such that∥C∥=1cm(n, n-1, ..., 2, 1)for each integer n>0 and how to construct such a family of min-wise independent permutations C for each integer n>0 if it exists. In this paper, we shall construct a family of permutations F_n for each integer n>0 and show that F_n is min-wise independent and ∥F_n∥=1cm(n, n-1, ..., 2, 1). Thus our construction of F_n is optimal in the sense of family size.

    CiNii Books

    researchmap

  • A Polynomial Time Sampling Algorithm for an Optimal Family of Min-Wise Independent Permutations (Models of Computation and Algorithms)

    Shinozaki Takahiro, Itoh Toshiya

    RIMS Kokyuroku   1093   74 - 80   1999.4

     More details

    Language:English   Publisher:Kyoto University  

    CiNii Books

    researchmap

▼display all

Awards

  • 情報・システムソサイエティ活動功労賞

    2018   電子情報通信学会  

     More details

  • Yamashita SIG Research Award

    2009  

     More details

    Country:Japan

    researchmap

  • The Awaya Prize from the Acoustical Society of Japan (ASJ)

    2008  

     More details

    Country:Japan

    researchmap

  • カナガワビエンナーレ 日本国際連合協会会長賞

    1987   神奈川県  

     More details

Research Projects

  • Stochastic analysis of microscopic earthquake interactions and physical understanding of earthquake source system

    Grant number:22K03753  2022.4 - 2026.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research  Grant-in-Aid for Scientific Research (C)

      More details

    Grant amount:\4290000 ( Direct Cost: \3300000 、 Indirect Cost:\990000 )

    researchmap

  • Spoken Language Acquisition Agent with Fluent Intonation

    Grant number:22K12069  2022.4 - 2025.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research  Grant-in-Aid for Scientific Research (C)

      More details

    Grant amount:\4160000 ( Direct Cost: \3200000 、 Indirect Cost:\960000 )

    researchmap

  • CEFR-Jに基づくCAN-DOタスク中心の教授と評価に関する総合的研究

    Grant number:20H00095  2020.4 - 2025.3

    日本学術振興会  科学研究費助成事業 基盤研究(A)  基盤研究(A)

    根岸 雅史, 投野 由紀夫, 奥村 学, 高田 智子, 片桐 徳昭, 中谷 安男, 能登原 祥之, 石井 康毅, 長沼 君主, 篠崎 隆宏, 工藤 洋路, 内田 諭, 村越 亮治, 大橋 由紀子, 和泉 絵美, 周 育佳

      More details

    Grant amount:\44720000 ( Direct Cost: \34400000 、 Indirect Cost:\10320000 )

    2020年度前半は研究チームの編成・計画の具体化と研究協力校の募集と依頼を行った。小中高と検討したが、最も可能性が高い京都府との連携を最初に模索し、CAN-DOリストを用いた CEFR-Jを基盤とする教育実践と評価を、高校レベルでは京都府立東舞鶴高等学校に研究協力校として受諾してもらい、詳細データ(短期・長期)を収集することになった。
    一方、具体的な授業への介入を行う以外に、全般的な CAN-DO 評価を CEFR-J CAN-DO テストを用いて実施する計画も立てられた。これに関しても、CEFR-J のメーリングリスト等で呼びかけて大規模に実施する予定であったが、2020年度後半からのコロナ感染拡大により、当初の予定通り学校募集等ができなくなった。
    またライティングのように大規模にデータ収集を不特定多数の学校で実施できる可能性も検討し、これに関してはさいたま市を対象に検討を進めていったが、こちらもコロナによる学校側の感染対策がさまざまな障害となり、十分に研究協力に時間を割くことが学校側としてできない状況があった。
    2020年度後半は予定を変更し、研究協力校に負担にならないように京都府の全体研修などの機会を利用して担当の教員と連絡を取り合い、こちら側の研究目的や教育支援体制を説明し、連携できる体制を整えることに時間を費やした。2020年度終盤に、次年度の予定を話し合い、まずは試験的に授業観察を行って授業データを録画・分析して、そこから課題を見いだして二学期に授業を焦点化して改善点を探ることとした。

    researchmap

  • Constraint Free Training of Speech Recognition Systems Based on Full Bayes Modeling

    Grant number:17K20001  2017.6 - 2020.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research  Grant-in-Aid for Challenging Research (Exploratory)

    Shinozaki Takahiro

      More details

    Grant amount:\6240000 ( Direct Cost: \4800000 、 Indirect Cost:\1440000 )

    The dependency on supervised learning using paired data is a major bottle-neck of current speech recognition systems. The goal of this research is to improve the flexibility of the system learning by using unpaired data. We have proposed a method to automatically extend the pronunciation dictionary from unmatched phoneme data and text data by applying the nonparametric Bayes method and weighted finite transducer. We have also worked on reinforcement learning of speech recognition systems by formulating the whole encoder-decoder based system as a policy function. We have shown that our proposed reinforcement learning methods significantly improve learning efficiency.

    researchmap

  • Research into CEFR-J-based 'can do' task and test development

    Grant number:16H01935  2016.4 - 2020.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research  Grant-in-Aid for Scientific Research (A)

    Negishi Masashi

      More details

    Grant amount:\37960000 ( Direct Cost: \29200000 、 Indirect Cost:\8760000 )

    The CEFR-J was a CEFR-based statistically validated framework for English language teaching in Japan. The purpose of the present study was to construct a battery of language tests to assess learners’ performance specified in “Can do” descriptors in the CEFR-J.
    For each of the five modes of communication, “Can do”-based performance tests were developed for Pre-A1 to B2.2 levels, with the help of CEFR-J lexical and grammatical profile information. In the final year, most performance test samples with validation reports were made publicly available at the CEFR-J official website, which will contribute to the promotion of “Can do”-based performance tests at school and the use of the CEFR as a reference tool.

    researchmap

  • Self-Organized Learning of Speech Recognition and Synthesis Systems

    Grant number:26280055  2014.4 - 2018.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)  Grant-in-Aid for Scientific Research (B)

    Shinozaki Takahiro, ARAI Takayuki, WATANABA Shinji, DUH Kevin

      More details

    Grant amount:\15730000 ( Direct Cost: \12100000 、 Indirect Cost:\3630000 )

    The purpose of this study is to make self-standing speech and language information processing systems that can learn from a small amount of labeled and a significant amount of unlabeled speech data as well as can automatically optimize its structure and learning conditions. We have proposed evolution strategy based automation method for neural network-based system development, series of semi-supervised learning methods for statistical speech models, and a reinforcement learning method of speech recognition systems. A high-performance Japanese speech recognition system integrating the research results have been published and widely used.

    researchmap

  • Practical application and validation of a computerized automatic scoring Japanese speaking test

    Grant number:26244026  2014.4 - 2017.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (A)  Grant-in-Aid for Scientific Research (A)

    IMAI Shingo, ISHIZUKA Kenkichi

      More details

    Grant amount:\37700000 ( Direct Cost: \29000000 、 Indirect Cost:\8700000 )

    We developed a testing system called SJ-CAT (Speaking Japanese Computerized Adaptive Test), which is accessible on the Internet. The test automatically measures the speaking ability of non-native speakers of Japanese language. SJ-CAT consists of four types of questions, i.e., reading a sentence, reading a correct sentence from three choices, making a sentence, and expressing one's opinion. The system evaluates one's speaking ability based on acoustic feature value (e.g. prosodic patterns, acoustic likelihood, and several kind of speaking rates) and keywords. Scores are calculated by means of a polytomous Item Response Model. Comparison between SJ-CAT and another speaking test, which is evaluated by trained human raters, showed high correlation, which indicates the practicality of SJ-CAT.

    researchmap

  • Speech information processing using deep generative models and their factorization

    Grant number:25280058  2013.4 - 2016.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)  Grant-in-Aid for Scientific Research (B)

    Shinoda Koichi, IWANO Koji, SHINOZAKI Takahiro

      More details

    Grant amount:\16900000 ( Direct Cost: \13000000 、 Indirect Cost:\3900000 )

    In speech recognition, it is important to train an accurate deep neural network (DNN) acoustic model from a large amount speech data from many speakers. In this study, we developed a framework to improve accuracy of the DNN acoustic model by factorizing speech data into phoneme and speaker elements. First we developed a speaker recognition method using deep Siamese network in which two DNNs which share its part. Second, we applied a DNN with a hierarchical phonetic structure to speaker adaptation. Third, we developed a speaker-adaptive training method where we utilized a student-teacher learning framework using soft targets. We improved speaker verification and speech recognition performance. We also studied DNN implementation and DNN structure design.

    researchmap

  • Macromolecular Potential Energy Decoder Based on Graphical Model

    Grant number:23650068  2011 - 2013

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Challenging Exploratory Research  Grant-in-Aid for Challenging Exploratory Research

    SHINOZAKI Takahiro, SHINODA Shinoda, SEKIJIMA Masakazu

      More details

    Grant amount:\3250000 ( Direct Cost: \2500000 、 Indirect Cost:\750000 )

    Knowing tertiary structure is important to understand and predict protein function. However, it is an open question how to predict the tertiary structure of proteins from a sequence of amino acids. In this project, Slice Chain Max-Sum (SCMS) algorithm has been proposed. This method represents the potential function of a protein molecule as a factor graph, which is a kind of a graphical model. The factor graph is converted into a linearly structured one according to a slicing of the molecule in 3D space. Based on the converted graph, max-sum search is performed in combination with node-wise local MCMC sampling that approximates continuous variables by discrete ones. Experimental results show that SCMS is more efficient than conventional MCMC method. It is also shown that improved version of SCMS (i.e. SCMS2.0) outperforms MCMC method that is reinforced by the quasi-Newton method.

    researchmap

  • Development of a Computer Automated Scoring Test of Spoken Japanese Using Speech Recognition Techniques

    Grant number:22242014  2010 - 2012

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (A)  Grant-in-Aid for Scientific Research (A)

    IMAI Shingo, ITO Sukero, NAKAMURA Yoichi, SAKAI Takako, AKAGI Yayoi, KIKUCHI Kenichi, HONDA Akiko, NAKASONO Hiromi, NISIMURA Ryuichi, SHINOZAKI Takahiro, YAMADA Takeshi, YANEHASHI Nobuko, ISHIZUKA Kenkichi, PHAM Thanh Son

      More details

    Grant amount:\46670000 ( Direct Cost: \35900000 、 Indirect Cost:\10770000 )

    We have developed a computer speaking test for Japanese learners, which automatically evaluates speaking ability on computers. It will be accessible on the internet anytime, anywhere. The automatic scoring system is implemented through speech recognition techniques, which obtains acoustic features from the utterance. The system is a computerized adaptive test based on Item Response Theory, which makes it possible to evaluate the speaking ability with relatively fewer test items by adjusting to the ability of test takers and to the difficulty of the test items.

    researchmap

  • 遅延評価手法を用いた大規模統計システム構築法の確立

    2010

      More details

    Grant type:Competitive

    researchmap

  • Robust Speaker Recognition with Intra-Speaker Variability Compensation based on Long-Term Recorded Speech Corpus

    Grant number:21300060  2009.4 - 2014.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)  Grant-in-Aid for Scientific Research (B)

    KUROIWA Shingo, TSUGE Satoru, OSANAI Takashi, SHINOZAKI Takahiro, HORIUCHI Yasuo, NISHIDA Masafumi

      More details

    Grant amount:\17940000 ( Direct Cost: \13800000 、 Indirect Cost:\4140000 )

    This research project aimed to build a new speech corpus that enables many researchers to investigate changes in human voices during a day, a month or several years, and to develop accurate and robust speaker recognition methods for industrial and forensic uses. The speech corpus named "AWA Long-Term Recorded Speech Corpus (AWA-LTR), which is released by Speech Resources Consortium of National Institute of Informatics (NII-SRC), consists of 6 speaker's read speech data recorded at morning, noon, and evening every week for several years (2 to 10 years). Using this corpus, we have developed intra-speaker variability compensation methods that improve the robustness of speaker recognition techniques. We also studied effective speech features for forensic speaker recognition, a comparison between human and machine speaker recognition abilities, accurate and robust speaker modeling methods and speaker verification methods.

    researchmap

  • Study on spoken language understanding framework integrating knowkedges among multiple layers

    Grant number:21300066  2009.4 - 2014.3

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)  Grant-in-Aid for Scientific Research (B)

    LEE Akinobu, KOMATANI Kazunori, NANJO Hiroaki, NISIMURA Ryuuichi, NISHIDA Masafumi, SHINOZAKI Takahiro, AKITA Yuya

      More details

    Grant amount:\17550000 ( Direct Cost: \13500000 、 Indirect Cost:\4050000 )

    This study focuses on developing a framework that integrates handling of multiple knowledge layer from speech signal processing to spoken language understanding directly into speech recognition process in a statistical mannar. Statistical models at layers of language model, acoustic model and dialogue model are widely investigated. For integration, speech decoding based on Bayes-risk minimization in which all the constraint can be expressed as Bayes risk, and some integration methods that utilizes speech information for dialogue management and turn taking was investigated. Part of the results are publicly available as part of an open-source voice interaction building tool MMDAgent and Julius.

    researchmap

  • Advancement of speech recognition technology using WFST

    Grant number:21300062  2009 - 2011

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Scientific Research (B)  Grant-in-Aid for Scientific Research (B)

    FURUI Sadaoki, SHINODA Koichi, SHINOZAKI Takahiro

      More details

    Grant amount:\18070000 ( Direct Cost: \13900000 、 Indirect Cost:\4170000 )

    With the aim of improving the performance of automatic speech recognition using the Weighted Finite State Transducer(WFST)-based decoder and developing new applications of the decoder, a wide range of research has been conducted and various achievements have been obtained. The world highest performance speech recognition decoder,"T^3 decoder", has been developed by improving the on-the-fly algorithm for the WFST decoder. Recognition performance under noisy environment has been improved by incorporating speech/non-speech information to the decoder. Various new techniques have been developed to apply the decoder to the recognition of resource-deficient languages and code-switching speech, and to transliteration. Innovative ideas have been proposed toward new directions of the decoder technology. T^3 decoder has been released to domestic as well as overseas research laboratories.

    researchmap

  • 目的音モデル尤度を用いた高速な耐雑音音声認識フロントエンドの研究

    2009 - 2011

      More details

    Grant type:Competitive

    researchmap

  • Efficient noise robust front-end based on target speech model likelihood for automatic speech recognition

    Grant number:21700188  2009 - 2010

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Young Scientists (B)  Grant-in-Aid for Young Scientists (B)

    SHINOZAKI Takahiro

      More details

    Grant amount:\4290000 ( Direct Cost: \3300000 、 Indirect Cost:\990000 )

    To improve speech recognition performance in adverse conditions, a noise compensation method is proposed and investigated that applies a transformation in the spectral domain whose parameters are optimized based on likelihood of speech GMM modeled on the feature domain. Experimental results show that the proposed method is able to work in real-time and it is effective to reduce noise effects.

    researchmap

  • CV 学習法を用いた最尤及び識別学習基準による準教師あり学習法の研究

    2009 - 2010

      More details

    Grant type:Competitive

    researchmap

  • Lightly supervised training based on CV framework using ML and discriminative criteria

    2009 - 2010

      More details

    Grant type:Competitive

    researchmap

  • Statistical pattern classifier training based on cross-validation likelihood

    2007 - 2009

      More details

    Grant type:Competitive

    researchmap

  • クロスバリデーション尤度を用いた統計的パターン分類器学習アルゴリズムの研究

    2007 - 2009

      More details

    Grant type:Competitive

    researchmap

  • Statistical pattern classifier training based on cross-validation likelihood

    Grant number:19700167  2007 - 2008

    Japan Society for the Promotion of Science  Grants-in-Aid for Scientific Research Grant-in-Aid for Young Scientists (B)  Grant-in-Aid for Young Scientists (B)

    SHINOZAKI Takahiro

      More details

    Grant amount:\3780000 ( Direct Cost: \3300000 、 Indirect Cost:\480000 )

    researchmap

▼display all