PARAMETERIZATIONOFPROSODICFEATUREDISTRIBUTIONSFOR SVM MODELING INSPEAKERRECOGNITION

Luciana Ferrer, Elizabeth Shriberg, Sachin Kajarekar

Abstract

Multiple recent studies haveshownthat speaker recognition performance using frame-based cepstral features isimproved byadding higher-level information, including prosodic andlexical features. Thispaper explores theimportant question offinding agoodkernelforasystem that models syllable-based prosodic features using support vector machines (SVMs). Thesystem hasbeenthebest performing ofourhigh-level systems inthelast twoNISTevaluations, andgives significant improvements whencombined withcepstralbased systems. Weintroduce twonewmethods fortransforming the syllable-level features into asingle high-dimensional vector that can bewell modeled bySVMs,resulting insignificant gains inspeaker recognition performance. IndexTerms-Speaker recognition, Prosody, GMM,SVM
BibTeX
@inproceedings{ferrer2007parameterizationofprosodicfeaturedistributionsfor,
  author = {Luciana Ferrer and Elizabeth Shriberg and Sachin Kajarekar},
  title = {{PARAMETERIZATIONOFPROSODICFEATUREDISTRIBUTIONSFOR SVM MODELING INSPEAKERRECOGNITION}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2007},
  url = {https://openreview.net/forum?id=fhKOO6ZO1WI},
}

Loading related content...

Figure