ISCA Archive Odyssey 2012
ISCA Archive Odyssey 2012

Online two speaker diarization

Hagai Aronowitz, Yosef A. Solewicz, Orith Toledo-Ronen

Short conversations pose some challenges for online diarization due to data sparseness and unbalanced representation of the two speakers. This paper presents our recent advances in online diarization of two-wire telephone conversations, introducing several methods for improving processing efficiency and accuracy on short conversations. Our framework is based on the offline diarization of a conversation prefix followed by an efficient online processing of the rest of the conversation. We use an adaptive prefix size, resulting from the tradeoff between desired efficiency and accuracy as measured by a confidence measure on the diarization output. We further show the enhancement of our online speaker recognition system based on implicit speaker diarization using the proposed techniques.

Cite as: Aronowitz, H., Solewicz, Y.A., Toledo-Ronen, O. (2012) Online two speaker diarization. Proc. The Speaker and Language Recognition Workshop (Odyssey 2012), 122-129

  author={Hagai Aronowitz and Yosef A. Solewicz and Orith Toledo-Ronen},
  title={{Online two speaker diarization}},
  booktitle={Proc. The Speaker and Language Recognition Workshop (Odyssey 2012)},