SPEAKER ADAPTATION USING EIGENVOICES
Robert Westwood
August 1999
The thesis considers a novel technique for adaptation of speaker models, called eigenvoice decomposition (ED), based around reducing the dimension of the search space of acoustic models. The technique is compared both practically and theoretically with several other adaptation techniques.
The use of Principal Component Analysis to choose the subspace is discussed and evaluated. One published method of choosing a model within this space, Maximal Likelihood Eigenvoice Decomposition, is presented and compared with a new method, Weighted Projection.
Robust estimation of the subspace was found to be difficult, but when it could be performed, ED was found to give a significant improvement in recognition accuracy with only one adaptation sentence, rather than the five or so required by other methods.
If you have difficulty viewing files that end '.gz',
which are gzip compressed, then you may be able to find
tools to uncompress them at the gzip
web site.
If you have difficulty viewing files that are in PostScript, (ending
'.ps' or '.ps.gz'), then you may be able to
find tools to view them at
the gsview
web site.
We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.