ISCA Archive AVIOS 2012
ISCA Archive AVIOS 2012

Optimizing feature representation for speaker diarization using PCA and LDA

Itshak Lapidot, Jean-Francois Bonastre

In this work we examine the interest of both LDA and PCA applied on the mel-cepstrum coefficients for speaker diarization. PCA is applied before the diarization process when LDA is used after an initial diarization step. We show that PCA allows a reduction in diarization time but do not offer a diarization error reduction contrarily to LDA which allows a performance improvement of about 14:8% (relative).


Cite as: Lapidot, I., Bonastre, J.-F. (2012) Optimizing feature representation for speaker diarization using PCA and LDA. Proc. Afeka-AVIOS Speech Processing Conference, 8-11

@inproceedings{lapidot12_avios,
  author={Itshak Lapidot and Jean-Francois Bonastre},
  title={{Optimizing feature representation for speaker diarization using PCA and LDA}},
  year=2012,
  booktitle={Proc. Afeka-AVIOS Speech Processing Conference},
  pages={8--11}
}