Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion

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

    Abstract

    A structure of Partly-Hidden Markov Model (PHMM) is optimized. PHMM was proposed in our previous work to deal with the complicated temporal changes of acoustic features. It can realize the observation dependent behaviors in both observations and state transitions. In the formulation of previous PHMM, we used a common structure in all model categories. However, it is well known that the optimal structure which gives best performance differs from category to category. In this paper, we designed a new structure optimization method in which the state-observation dependences in PHMM are optimally defined with respect to each category using Weighted Likelihood-Ratio Maximization (WLRM) criterion. WLRM criterion induces sparse and discriminative structures, and therefore gives the resulting structurally discriminative models. We define the model structure combination which gives maximum weighted likelihood-ratio for any possible structure patterns as the optimal structures, and Genetic Algorithm is applied to an optimal approximation of search. As the results of continuous speech recognition aiming at lecture talks, the effectiveness of the proposed structure optimization is shown: it reduced the word errors compared to HMM and PHMM with common structure for all categories.

    Original languageEnglish
    Title of host publication9th European Conference on Speech Communication and Technology
    Pages3353-3356
    Number of pages4
    Publication statusPublished - 2005
    Event9th European Conference on Speech Communication and Technology - Lisbon
    Duration: 2005 Sep 42005 Sep 8

    Other

    Other9th European Conference on Speech Communication and Technology
    CityLisbon
    Period05/9/405/9/8

    Fingerprint

    Hidden Markov models
    Continuous speech recognition
    Model structures
    Maximum likelihood
    Genetic algorithms
    Acoustics

    ASJC Scopus subject areas

    • Engineering(all)

    Cite this

    Ogawa, T., & Kobayashi, T. (2005). Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion. In 9th European Conference on Speech Communication and Technology (pp. 3353-3356)

    Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion. / Ogawa, Tetsuji; Kobayashi, Tetsunori.

    9th European Conference on Speech Communication and Technology. 2005. p. 3353-3356.

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

    Ogawa, T & Kobayashi, T 2005, Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion. in 9th European Conference on Speech Communication and Technology. pp. 3353-3356, 9th European Conference on Speech Communication and Technology, Lisbon, 05/9/4.
    Ogawa T, Kobayashi T. Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion. In 9th European Conference on Speech Communication and Technology. 2005. p. 3353-3356
    @inproceedings{4cc1277f46b0452b88060612cdd861b0,
    title = "Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion",
    abstract = "A structure of Partly-Hidden Markov Model (PHMM) is optimized. PHMM was proposed in our previous work to deal with the complicated temporal changes of acoustic features. It can realize the observation dependent behaviors in both observations and state transitions. In the formulation of previous PHMM, we used a common structure in all model categories. However, it is well known that the optimal structure which gives best performance differs from category to category. In this paper, we designed a new structure optimization method in which the state-observation dependences in PHMM are optimally defined with respect to each category using Weighted Likelihood-Ratio Maximization (WLRM) criterion. WLRM criterion induces sparse and discriminative structures, and therefore gives the resulting structurally discriminative models. We define the model structure combination which gives maximum weighted likelihood-ratio for any possible structure patterns as the optimal structures, and Genetic Algorithm is applied to an optimal approximation of search. As the results of continuous speech recognition aiming at lecture talks, the effectiveness of the proposed structure optimization is shown: it reduced the word errors compared to HMM and PHMM with common structure for all categories.",
    author = "Tetsuji Ogawa and Tetsunori Kobayashi",
    year = "2005",
    language = "English",
    pages = "3353--3356",
    booktitle = "9th European Conference on Speech Communication and Technology",

    }

    TY - GEN

    T1 - Optimizing the structure of partly-hidden Markov models using weighted likelihood-ratio maximization criterion

    AU - Ogawa, Tetsuji

    AU - Kobayashi, Tetsunori

    PY - 2005

    Y1 - 2005

    N2 - A structure of Partly-Hidden Markov Model (PHMM) is optimized. PHMM was proposed in our previous work to deal with the complicated temporal changes of acoustic features. It can realize the observation dependent behaviors in both observations and state transitions. In the formulation of previous PHMM, we used a common structure in all model categories. However, it is well known that the optimal structure which gives best performance differs from category to category. In this paper, we designed a new structure optimization method in which the state-observation dependences in PHMM are optimally defined with respect to each category using Weighted Likelihood-Ratio Maximization (WLRM) criterion. WLRM criterion induces sparse and discriminative structures, and therefore gives the resulting structurally discriminative models. We define the model structure combination which gives maximum weighted likelihood-ratio for any possible structure patterns as the optimal structures, and Genetic Algorithm is applied to an optimal approximation of search. As the results of continuous speech recognition aiming at lecture talks, the effectiveness of the proposed structure optimization is shown: it reduced the word errors compared to HMM and PHMM with common structure for all categories.

    AB - A structure of Partly-Hidden Markov Model (PHMM) is optimized. PHMM was proposed in our previous work to deal with the complicated temporal changes of acoustic features. It can realize the observation dependent behaviors in both observations and state transitions. In the formulation of previous PHMM, we used a common structure in all model categories. However, it is well known that the optimal structure which gives best performance differs from category to category. In this paper, we designed a new structure optimization method in which the state-observation dependences in PHMM are optimally defined with respect to each category using Weighted Likelihood-Ratio Maximization (WLRM) criterion. WLRM criterion induces sparse and discriminative structures, and therefore gives the resulting structurally discriminative models. We define the model structure combination which gives maximum weighted likelihood-ratio for any possible structure patterns as the optimal structures, and Genetic Algorithm is applied to an optimal approximation of search. As the results of continuous speech recognition aiming at lecture talks, the effectiveness of the proposed structure optimization is shown: it reduced the word errors compared to HMM and PHMM with common structure for all categories.

    UR - http://www.scopus.com/inward/record.url?scp=33745193680&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=33745193680&partnerID=8YFLogxK

    M3 - Conference contribution

    SP - 3353

    EP - 3356

    BT - 9th European Conference on Speech Communication and Technology

    ER -