Sixth International Conference on Spoken Language Processing
In this paper a method is presented to estimate a broad phonetic class regression tree to be used in MLLR adap- tation. The tree is derived from the correlation structure among phone units estimated on the training data. The al- gorithm is language-independent and showed good results on both an English and a Mandarin Chinese database. In adaptation experiments the tree outperformed a regres- sion tree obtained from clustering according to closeness in acoustic space and achieved results comparable with those of a manually designed broad phonetic class tree.
Bibliographic reference. Haeb-Umbach, Reinhold (2000): "Data-driven phonetic regression class tree estimation for MLLR adaptation", In ICSLP-2000, vol.3, 857-860.