COST278 and ISCA Tutorial and Research Workshop (ITRW) on Robustness Issues in Conversational Interaction

University of East Anglia, Norwich, UK
August 30-31, 2004

Crosslingual Transfer of Source Acoustic Models to Two Different Target Languages

Andrej Zgank (1), Zdravko Kacic (1), Klara Vicsi (2), Gyorgy Szaszak (2), Frank Diehl (3), Jozef Juhar (4), Slavomir Lihan (4)

(1) University of Maribor, Maribor, Slovenia
(2) Budapest University of Technology and Economics, Budapest, Hungary
(3) Universitat Politècnica de Catalunya, Barcelona, Spain
(4) Technical University of Kosice, Kosice, Slovakia

This paper presents the ongoing work on crosslingual speech recognition in the MASPER initiative. Source acoustic models were transferred to two different target languages - Hungarian and Slovenian. Beside the monolingual source acoustic models, also a semi-multilingual set was defined. An expert-knowledge approach and a data-driven method were applied for transfer. The crosslingual speech recognition results were used to analyse the robustness of different source acoustic models respective the language similarity influence.

Full Paper

Bibliographic reference.  Zgank, Andrej / Kacic, Zdravko / Vicsi, Klara / Szaszak, Gyorgy / Diehl, Frank / Juhar, Jozef / Lihan, Slavomir (2004): "Crosslingual transfer of source acoustic models to two different target languages", In Robust2004, paper 19.