[Subject Prev][Subject Next][Thread Prev][Thread Next][Date Index][Thread Index]

[hts-users:03330] Re: About the Monophone Models

On 6 Jun 2012, at 10:07, Kwan Lisa wrote:

> I'm wondering why this matrix was not trained.

because there are no trainable transition parameters in this model topology

you have created a model with exactly one transition out of each state: therefore the probability on all transitions is 1 by definition

(note: a zero-probability transition will always remain zero after training - read up on the Baum-Welch algorithm in a textbook to work out why)


The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.

[hts-users:03331] Re: About the Monophone Models, Simon King
[hts-users:03329] About the Monophone Models, Kwan Lisa