Cross-Dialectal Data Transferring for Gaussian Mixture Model Training in Arabic Speech Recognition
Po-Sen Huang, Mark Hasegawa‐Johnson · 2012
Abstract—Dialectal Arabic speech recognition is a difficult problem and is relatively less studied. In this paper, we propose a cross-dialectal Gaussian mixture model training criteria to transfer knowledge from one domain to the other by data sharing. Specifically, phone classification experiments on West Point Modern Standard Arabic Speech corpus and Babylon Levantine Arabic Speech corpus demonstrate that the cross-dialectal training improves phone classification accuracy significantly, especially when a small amount of MSA data is transferred. Keywords—Transfer learning, Arabic automatic speech recognition, Gaussian Mixture Models, dialectal Arabic