Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Combining the glottal mixture model (GLOMM) with UBM for speaker recognition

: Baggenstoss, P.M.


Institute of Electrical and Electronics Engineers -IEEE-; European Association for Signal Processing -EURASIP-:
24th European Signal Processing Conference, EUSIPCO 2016. Proceedings : 28 August - 2 September 2016, Budapest, Hungary
Piscataway, NJ: IEEE, 2016
ISBN: 978-1-5090-1891-8 (Print)
ISBN: 978-0-9928-6265-7 (Online)
European Signal Processing Conference (EUSIPCO) <24, 2016, Budapest>
Conference Paper
Fraunhofer FKIE ()

We present an iterative algorithm to extract the voice source waveform from recordings of speech for speaker identification. The method detects glottal closings, then constructs a speaker-dependent library of glottal pulse waveforms by clustering data windows centered on the linear prediction error time-series at the glottal closures. With the voice source modeled as scaled and shifted glottal pulses, the algorithm iteratively determines the vocal tract parameters in each frame. In experiments, we combine the extracted voice source information with a universal background model (UBM). Using the TIMIT data corpus and a 200-speaker population size, we demonstrate a factor of three speaker recognition error reduction.