Integration of MKL-based and I-vector-based speaker verification by short utterances

Hideitsu Hino, Tetsuji Ogawa

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    We developed a speaker verification system that is efficient for short utterances. The i-vector-based speaker representation has helped realize highly accurate speaker verification systems, however, it might be not robust against short utterances because the reliability of statistics required for extracting i-vectors is low. On the other hand, multiple kernel learning based on conditional entropy minimization has also achieved high accuracy in speaker verification that is robust against intra-speaker variability. To improve the robustness of speaker verification systems against short utterances, we attempted to integrate the above-mentioned complementary systems. Our experimental results showed that the proposed system integration achieved high-accuracy speaker verification systems, irrespective of the utterance lengths, even for very short utterances (e.g., less than two seconds).

    Original languageEnglish
    Title of host publicationProceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013
    PublisherIEEE Computer Society
    Pages562-566
    Number of pages5
    DOIs
    Publication statusPublished - 2013
    Event2013 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013 - Naha, Okinawa
    Duration: 2013 Nov 52013 Nov 8

    Other

    Other2013 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013
    CityNaha, Okinawa
    Period13/11/513/11/8

    Fingerprint

    Entropy
    Statistics

    Keywords

    • I-vector
    • Multiple kernel learning
    • Speaker verification

    ASJC Scopus subject areas

    • Computer Vision and Pattern Recognition

    Cite this

    Hino, H., & Ogawa, T. (2013). Integration of MKL-based and I-vector-based speaker verification by short utterances. In Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013 (pp. 562-566). [6778381] IEEE Computer Society. https://doi.org/10.1109/ACPR.2013.42

    Integration of MKL-based and I-vector-based speaker verification by short utterances. / Hino, Hideitsu; Ogawa, Tetsuji.

    Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013. IEEE Computer Society, 2013. p. 562-566 6778381.

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Hino, H & Ogawa, T 2013, Integration of MKL-based and I-vector-based speaker verification by short utterances. in Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013., 6778381, IEEE Computer Society, pp. 562-566, 2013 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013, Naha, Okinawa, 13/11/5. https://doi.org/10.1109/ACPR.2013.42
    Hino H, Ogawa T. Integration of MKL-based and I-vector-based speaker verification by short utterances. In Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013. IEEE Computer Society. 2013. p. 562-566. 6778381 https://doi.org/10.1109/ACPR.2013.42
    Hino, Hideitsu ; Ogawa, Tetsuji. / Integration of MKL-based and I-vector-based speaker verification by short utterances. Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013. IEEE Computer Society, 2013. pp. 562-566
    @inproceedings{7ad6697eae20456db10eb6346a654c45,
    title = "Integration of MKL-based and I-vector-based speaker verification by short utterances",
    abstract = "We developed a speaker verification system that is efficient for short utterances. The i-vector-based speaker representation has helped realize highly accurate speaker verification systems, however, it might be not robust against short utterances because the reliability of statistics required for extracting i-vectors is low. On the other hand, multiple kernel learning based on conditional entropy minimization has also achieved high accuracy in speaker verification that is robust against intra-speaker variability. To improve the robustness of speaker verification systems against short utterances, we attempted to integrate the above-mentioned complementary systems. Our experimental results showed that the proposed system integration achieved high-accuracy speaker verification systems, irrespective of the utterance lengths, even for very short utterances (e.g., less than two seconds).",
    keywords = "I-vector, Multiple kernel learning, Speaker verification",
    author = "Hideitsu Hino and Tetsuji Ogawa",
    year = "2013",
    doi = "10.1109/ACPR.2013.42",
    language = "English",
    pages = "562--566",
    booktitle = "Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013",
    publisher = "IEEE Computer Society",

    }

    TY - GEN

    T1 - Integration of MKL-based and I-vector-based speaker verification by short utterances

    AU - Hino, Hideitsu

    AU - Ogawa, Tetsuji

    PY - 2013

    Y1 - 2013

    N2 - We developed a speaker verification system that is efficient for short utterances. The i-vector-based speaker representation has helped realize highly accurate speaker verification systems, however, it might be not robust against short utterances because the reliability of statistics required for extracting i-vectors is low. On the other hand, multiple kernel learning based on conditional entropy minimization has also achieved high accuracy in speaker verification that is robust against intra-speaker variability. To improve the robustness of speaker verification systems against short utterances, we attempted to integrate the above-mentioned complementary systems. Our experimental results showed that the proposed system integration achieved high-accuracy speaker verification systems, irrespective of the utterance lengths, even for very short utterances (e.g., less than two seconds).

    AB - We developed a speaker verification system that is efficient for short utterances. The i-vector-based speaker representation has helped realize highly accurate speaker verification systems, however, it might be not robust against short utterances because the reliability of statistics required for extracting i-vectors is low. On the other hand, multiple kernel learning based on conditional entropy minimization has also achieved high accuracy in speaker verification that is robust against intra-speaker variability. To improve the robustness of speaker verification systems against short utterances, we attempted to integrate the above-mentioned complementary systems. Our experimental results showed that the proposed system integration achieved high-accuracy speaker verification systems, irrespective of the utterance lengths, even for very short utterances (e.g., less than two seconds).

    KW - I-vector

    KW - Multiple kernel learning

    KW - Speaker verification

    UR - http://www.scopus.com/inward/record.url?scp=84899120837&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=84899120837&partnerID=8YFLogxK

    U2 - 10.1109/ACPR.2013.42

    DO - 10.1109/ACPR.2013.42

    M3 - Conference contribution

    AN - SCOPUS:84899120837

    SP - 562

    EP - 566

    BT - Proceedings - 2nd IAPR Asian Conference on Pattern Recognition, ACPR 2013

    PB - IEEE Computer Society

    ER -