The goal of this paper is to introduce an integrated approach for speech and speaker recognition. A speaker identification device is integrated into an existing speech recognizer. The interaction of speech and speaker recognition enhances both recognition rates concerning speech and speaker identity as speech recognition exploits the knowledge of the speaker and vice versa. The speaker identification supports the speech recognizer to build up speaker-dependent models for speech recognition and speaker identification by unsupervised online speaker adaptation starting from a speaker-independent codebook. Thus speaker adaptation is enabled to adapt to different speakers which results in an optimal longterm adaptation. Unlike existing speaker identification systems this approach requires only a short training phase for new speakers. During the speech decoding process the identity of the current speaker is estimated and the speaker dependent codebook is further adapted after each identified utterance. Thus the speech controlled system has to handle speaker changes, speaker identification and speaker adaptation in an unsupervised manner. Furthermore the system accounts for different training levels of the statistical models by providing a smooth transition between short-term and long-term speaker adaptation depending on the amount of speakerdependent utterances. In the following a brief introduction to speaker identification with Gaussian Mixture Models (GMM) and speech recognition based on semi-continuous Hidden Markov Models (HMM) is given. Fast and long-term speaker adaptation are briefly described and the proposed combination is presented. Then the new joint speech and speaker recognition system is described. The system is evaluated in an realistic in-car environment for a limited number of enrolled speakers. Speech and speaker recognition rates are presented for a subset of the Speecon database.


    Zugriff

    Zugriff über TIB

    Verfügbarkeit in meiner Bibliothek prüfen


    Exportieren, teilen und zitieren



    Titel :

    Joint speaker identification and speech recognition for speech controlled applications in an automotive environment


    Beteiligte:
    Herbig, T. (Autor:in) / Gerl, F. (Autor:in)


    Erscheinungsdatum :

    2009


    Format / Umfang :

    4 Seiten, 2 Bilder, 1 Tabelle, 7 Quellen


    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Print


    Sprache :

    Englisch




    Mathematical Analysis and Speaker-Independent Speech Recognition

    Gur'yanov, A. E. | British Library Online Contents | 1996


    Development of multi-lingual speech recognition and texted speech synthesis for automotive applications

    Deguchi, Y. / Kagoshima, T. / Hirabayashi, G. et al. | Tema Archiv | 2003


    Development of multi-lingual speech recognition and text-to speech synthesis for automotive applications

    Deguchi, Y. / Kagoshima, T. / Hirabayashi, G. et al. | British Library Conference Proceedings | 2003


    Development of multi-lingual speech recognition and text-to speech synthesis for automotive applications

    Deguchi,T. / Kagoshima,T. / Hirabayashi,G. et al. | Kraftfahrwesen | 2003


    Development of Multi-Lingual Speech Recognition and Text-to-Speech Synthesis for Automotive Applications

    Deguchi, Y. / Kagoshima, T. / Hirabayashi, G. et al. | British Library Conference Proceedings | 2002