| 電子情報通信学会 研究会発表申込システム 講演論文 詳細 |
技報閲覧サービス [ログイン] 技報アーカイブ |
| トップに戻る | 前のページに戻る | [Japanese] / [English] |
| 講演抄録／キーワード | ||
| 講演名 | 2008-03-20 15:15 ［ポスター講演］A Fast Sequence Kernel for Sequential Data Classification ○Makoto Yamada（SOKENDAI/Yamaha Corp.）・Tomoko Matsui（Inst. of Statistical Mathematics）  SP2007-199 |
|
| 抄録 | （和） | Sequence kernel based speaker verification systems have been shown to outperform the conventional systems based on Gaussian mixture models(GMMs) and support vector machines(SVMs) with vector kernels[1]. However, sequence kernels have two significant problems. First, sequence kernels tend to produce a diagonal dominant kernel Gram matrix i.e., only the diagonal elements of the Gram matrix have some large values. Second, the computation of sequence kernels requires $m \times n$ vector kernel computation for measuring the similarity between $m$ and $n$ lengths of sequential data. In this paper, we first study log normalization of a sequence kernel to avoid the diagonal dominance problem, and then address the fast implementation of sequential kernels by solving the pre-image problem. In text-independent speaker identification experiments with 10 male speakers, our approach was not only found to be competitive in identification rates with the conventional sequence kernel, but also achieved two to ten time faster sequence kernel computation. [1] J. Mariethoz and S. Bengio, ``A kernel trick for sequences applied to text-independent speaker verification systems,'' Pattern Recognition, 40, 2315-2324, 2007. |
| （英） | Sequence kernel based speaker verification systems have been shown to outperform the conventional systems based on Gaussian mixture models(GMMs) and support vector machines(SVMs) with vector kernels[1]. However, sequence kernels have two significant problems. First, sequence kernels tend to produce a diagonal dominant kernel Gram matrix i.e., only the diagonal elements of the Gram matrix have some large values. Second, the computation of sequence kernels requires $m \times n$ vector kernel computation for measuring the similarity between $m$ and $n$ lengths of sequential data. In this paper, we first study log normalization of a sequence kernel to avoid the diagonal dominance problem, and then address the fast implementation of sequential kernels by solving the pre-image problem. In text-independent speaker identification experiments with 10 male speakers, our approach was not only found to be competitive in identification rates with the conventional sequence kernel, but also achieved two to ten time faster sequence kernel computation. [1] J. Mariethoz and S. Bengio, ``A kernel trick for sequences applied to text-independent speaker verification systems,'' Pattern Recognition, 40, 2315-2324, 2007. |
|
| キーワード | （和） | Sequence kernel / pre-image problem / kernel logistic regression / speaker identification / / / / |
| （英） | Sequence kernel / pre-image problem / kernel logistic regression / speaker identification / / / / | |
| 文献情報 | 信学技報, vol. 107, no. 551, SP2007-199, pp. 75-80, 2008年3月. | |
| 資料番号 | SP2007-199 | |
| 発行日 | 2008-03-13 (SP) | |
| ISSN | Print edition: ISSN 0913-5685 Online edition: ISSN 2432-6380 | |
| 著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します．(許諾番号：10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) | |
| PDFダウンロード | SP2007-199 |
|
| 研究会情報 | |
| 研究会 | SP |
| 開催期間 | 2008-03-20 - 2008-03-21 |
| 開催地（和） | 東大 |
| 開催地（英） | Univ. Tokyo |
| テーマ（和） | 国際ワークショップ(20日），研究会(21日) 「聴覚・音声・言語とその障害，一般」 |
| テーマ（英） | International Workshop (Mar 20), Speech Production, Speech Perception, Hearing and Speech, etc. (Mar 21) |
| 講演論文情報の詳細 | |
| 申込み研究会 | SP |
| 会議コード | 2008-03-SP |
| 本文の言語 | 英語 |
| タイトル（和） | |
| サブタイトル（和） | |
| タイトル（英） | A Fast Sequence Kernel for Sequential Data Classification |
| サブタイトル（英） | |
| キーワード(1)（和/英） | Sequence kernel / Sequence kernel |
| キーワード(2)（和/英） | pre-image problem / pre-image problem |
| キーワード(3)（和/英） | kernel logistic regression / kernel logistic regression |
| キーワード(4)（和/英） | speaker identification / speaker identification |
| キーワード(5)（和/英） | / |
| キーワード(6)（和/英） | / |
| キーワード(7)（和/英） | / |
| キーワード(8)（和/英） | / |
| 第1著者 氏名（和/英/ヨミ） | Makoto Yamada / Makoto Yamada / |
| 第1著者 所属（和/英） | Graduate School for Advanced Studies/Yamaha Corporation (略称： 総研大/ヤマハ) Graduate School for Advanced Studies/Yamaha Corporation (略称： SOKENDAI/Yamaha Corp.) |
| 第2著者 氏名（和/英/ヨミ） | Tomoko Matsui / Tomoko Matsui / |
| 第2著者 所属（和/英） | Institute for Statistical Mathematics (略称： Inst. of Statistical Mathematics) Institute for Statistical Mathematics (略称： Inst. of Statistical Mathematics) |
| 講演者 | 第1著者 |
| 発表日時 | 2008-03-20 15:15:00 |
| 発表時間 | 90分 |
| 申込先研究会 | SP |
| 資料番号 | SP2007-199 |
| 巻番号（vol） | vol.107 |
| 号番号（no） | no.551 |
| ページ範囲 | pp.75-80 |
| ページ数 | 6 |
| 発行日 | 2008-03-13 (SP) |
[研究会発表申込システムのトップページに戻る]