University of Birmingham > Talks@bham > Speech Recognition by Synthesis Seminars > Baum-Welch re-estimation for some simple CSHMMs

Baum-Welch re-estimation for some simple CSHMMs

Add to your list(s) Download to your calendar using vCal

  • UserColin Champion, University of Birmingham
  • ClockWednesday 17 February 2016, 14:00-15:00
  • HouseGisbert Kapp, N123.

If you have a question about this talk, please contact Dr. Philip Weber.

I shall discuss parameter estimation for two simple forms of CSHMM - the dwell-only model and the sloping loudness model. The former is equivalent (for estimation purposes) to the hierarchical Gaussian model. I will sketch some of the literature on the topic, compare the Baum-Welch algorithm with the EM algorithm, show how Baum-Welch estimation can be performed, and remark on the fact that the mathematics is equivalent to the theory underlying maximum likelihood estimation of Gaussian parameters.

The sloping loudness model can be approached in much the same way. The mathematics becomes more complicated if realisations occur for which there is only one observation: in this case I show that the mathematics is equivalent to a generalisation of the earlier problem which has been studied by Anderson and Olkin.

I shall present results on TIMIT consonants if I have any.

This talk is part of the Speech Recognition by Synthesis Seminars series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


Talks@bham, University of Birmingham. Contact Us | Help and Documentation | Privacy and Publicity.
talks@bham is based on from the University of Cambridge.