dear sir,
when i run training with lda/mllt enable, i get error can not create lda file
. then i try to debug lda.py script focus on this routine:
If CMN was used, this should actually be very close to zero
globalmean = mean.sum(0) / dnom.sum()
sw = var.sum(0)
sb = numpy.zeros(var.shape, 'd')
for d, m in itertools.izip(dnom, mean):
diff = m / d - globalmean
sb += d * numpy.outer(diff, diff)
i trace value of "d" and found that sometimes it get zero value , that is the
reason why lda.py can not create lda file.
could you tell me why d get zero value and how to fix it ?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
dear sir,
when i run training with lda/mllt enable, i get error can not create lda file
. then i try to debug lda.py script focus on this routine:
If CMN was used, this should actually be very close to zero
globalmean = mean.sum(0) / dnom.sum()
sw = var.sum(0)
sb = numpy.zeros(var.shape, 'd')
for d, m in itertools.izip(dnom, mean):
diff = m / d - globalmean
sb += d * numpy.outer(diff, diff)
i trace value of "d" and found that sometimes it get zero value , that is the
reason why lda.py can not create lda file.
could you tell me why d get zero value and how to fix it ?
You have unsufficient data for particular phone
Add more training data or use less phones
And of course check that most your utterances aren't rejected due to not
reaching final state. You can check this in the logs in logdir subfolder.