Speaker Adaptation Based on PARAFAC2 of Transformation Matrices for Continuous Speech Recognition

Yongwon JEONG  Sangjun LIM  Young Kuk KIM  Hyung Soon KIM  

Publication
IEICE TRANSACTIONS on Information and Systems   Vol.E96-D   No.9   pp.2152-2155
Publication Date: 2013/09/01
Online ISSN: 1745-1361
DOI: 10.1587/transinf.E96.D.2152
Print ISSN: 0916-8532
Type of Manuscript: LETTER
Category: Speech and Hearing
Keyword: 
maximum likelihood linear regression,  parallel factor analysis,  PARAFAC2,  speaker adaptation,  speech recognition,  

Full Text: PDF(135.3KB)
>>Buy this Article


Summary: 
We present an acoustic model adaptation method where the transformation matrix for a new speaker is given by the product of bases and a weight matrix. The bases are built from the parallel factor analysis 2 (PARAFAC2) of training speakers' transformation matrices. We perform continuous speech recognition experiments using the WSJ0 corpus.