Speaker Recognition using a Non - parametric Speaker Model Representation and Earth Mover's Distance

• Item type: SIG Technical Reports(1)
• 公開日: 2004-12-20
• タイトル: Speaker Recognition using a Non - parametric Speaker Model Representation and Earth Mover's Distance
• タイトル / 言語: en
• タイトル: Speaker Recognition using a Non - parametric Speaker Model Representation and Earth Mover's Distance
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Faculty of Engineering Tokushima University
• 著者所属: Faculty of Engineering Tokushima University
• 著者所属: Faculty of Engineering Tokushima University
• 著者所属: Faculty of Engineering Tokushima University
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering, Tokushima University
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering, Tokushima University
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering, Tokushima University
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering, Tokushima University
• 著者名: Yoshiyuki, UMEDA; Satoru, TSUGE; Fuji, REN; Shingo, KUROIWA
• 著者名(英): Yoshiyuki, Umeda; Satoru, Tsuge; Fuji, Ren; ShingoKuroiwa
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: In this paper we propose a distributed speaker recognition method using a non-parametric speaker model and Earth Movers' Distance (EMD). In distributed speaker recognition the quantized feature vectors are sent to a server. The Gaussian mixture model (GMM) the traditional method used for speaker recognition is trained using the maximum likehood approach. However it is difficult to fit continuous density functions to quantized data. To overcome this problem the proposed method represents each speaker model with a speaker-dependent VQ code histogram designed by registered feature vectores and directly calculated the distance between the histograms of speaker models and testing quantized feature vectores. To measure the distance between each speaker model and testing data we use EMD which can calculate the distance between histograms with different bins. We conducted text-independent speaker identification experiments using the proposed method. Compared to results using the traditional GMM the proposed method yielded relative error reductions of 32% for quantized data.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: In this paper, we propose a distributed speaker recognition method using a non-parametric speaker model and Earth Movers' Distance (EMD). In distributed speaker recognition, the quantized feature vectors are sent to a server. The Gaussian mixture model (GMM), the traditional method used for speaker recognition, is trained using the maximum likehood approach. However, it is difficult to fit continuous density functions to quantized data. To overcome this problem, the proposed method represents each speaker model with a speaker-dependent VQ code histogram designed by registered feature vectores and directly calculated the distance between the histograms of speaker models and testing quantized feature vectores. To measure the distance between each speaker model and testing data, we use EMD which can calculate the distance between histograms with different bins. We conducted text-independent speaker identification experiments using the proposed method. Compared to results using the traditional GMM, the proposed method yielded relative error reductions of 32% for quantized data.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10442647
• 書誌情報: 情報処理学会研究報告音声言語情報処理（SLP）; 巻 2004, 号 131(2004-SLP-054), p. 85-90, 発行日 2004-12-20
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会