VSE knjižnice (vzajemna bibliografsko-kataložna baza podatkov COBIB.SI)
  • Speaker normalization in articulation training systems using analysis by synthesis
    Ogner, Marcel ; Kačič, Zdravko
    In this paper we present the spectrum normalization algorithms we implemented ina teaching and training system for hearing handicapped children, with the goal of more uniform visual representation of ... speech characteristics to enableefficient training and accurate evaluation of correct pronunciation of vowels and fricatives. In order to perform the vocal tract length normalization and excitation spectrum normalization simultaneously, the new speech model is needed that would enable us the spectrum decomposition into spectral envelope and excitaiton model. The model has to be able of spectum resynthesis after some sort of manipulation on both components. The vocal tract normalization is deduced by spectral envelope warping according to the frequency warping function extracted from the formant patterns for both speakers, reference speaker and tst speaker. The frequency warping has to be locally constarined to preserve all important information of poor articulation.The excitation spectrum normalization is carries out by talking the periodic and noisy component of excitation from the reference speaker. Both components are than mixed together according to the frequency dependent V/UV mixture function etimated for the test speaker.
    Vir: Advances in speech technology : seventh International Workshop, Maribor, Slovenia, July 4-5, 2000 : proceedings (Str. 213-218)
    Vrsta gradiva - prispevek na konferenci
    Leto - 2001
    Jezik - angleški
    COBISS.SI-ID - 6310422

Vnos na polico