User profiling through spoken utterances for user specific service

Minseok Keum, David K. Han, Hanseok Ko

    Research output: Contribution to conferencePaperpeer-review

    Abstract

    For providing user specific services in vehicle environment, we propose a speech based interface that identifies gender, age, and language of a user during speech utterances. For gender and age recognition, physiological factors are important discriminative information. However, for language recognition, phonotactics in addition to spectral information is exploited. All of the above schemes are evaluated using Mel-Frequency Cepstral Coefficient (MFCC) based features and Gaussian Mixture Models (GMM). We also evaluate the effectiveness of channel compensation techniques such as Cepstral Mean Subtraction (CMS) and RASTA filtering. As a conclusion we suggest that MFCC based feature can effectively be applied to the speech related recognition problems.

    Original languageEnglish
    Publication statusPublished - 2011
    Event5th Biennial Workshop on Digital Signal Processing for In-Vehicle Systems, DSP 2011 - Kiel, Germany
    Duration: 2011 Sept 42011 Sept 7

    Other

    Other5th Biennial Workshop on Digital Signal Processing for In-Vehicle Systems, DSP 2011
    Country/TerritoryGermany
    CityKiel
    Period11/9/411/9/7

    Bibliographical note

    Funding Information:
    This research work was supported in part by CNPq and CAPES Brazilian funding agencies and UGC-BSR Research Start-Up-Grant F.30-420/2018(BSR), UGC, India. M. Chand would like thank CSIR-NPL, New Delhi, India for using magneto-rheometer to carry out rheological measurements.

    Keywords

    • Cepstral-Time Matrix (CTM)
    • Channel compensation
    • Gaussian Mixture Models (GMM)
    • Mel-Frequency Cepstral Coefficient (MFCC)
    • Speech user profiling

    ASJC Scopus subject areas

    • Signal Processing
    • Automotive Engineering

    Fingerprint

    Dive into the research topics of 'User profiling through spoken utterances for user specific service'. Together they form a unique fingerprint.

    Cite this