Interspeech'2005 - Eurospeech
This paper proposes a method for automatically clustering multilingual speech so as to derive language family trees. We consider that the language is the source of information which generates speech feature parameters; the probability or statistical characteristics of this information is modeled by Gaussian mixture models (GMMs); then a distance measure between the GMMs is introduced. Based on this, we construct family trees of multilingual speech which are quite similar to those considered in linguistics.
Bibliographic reference. Itahashi, Shuichi / Zhu, Shiwei / Yamamoto, Mikio (2005): "Constructing family trees of multilingual speech using Gaussian mixture models", In INTERSPEECH-2005, 2253-2256.