Javi,
Thank you for your response.
How much data (~1000 samples, ~10000 samples, more) may be
sufficient in order to have robust models without HINIT, HEREST?
Can you please provide me more information?
Best Regards,
Stas
PS. Sébastien, thank you for your help.
On Sun, Aug 9, 2009 at 12:16 PM, Xavi Gonzalvo
<xavi.gonzalvo@xxxxxxxxx> wrote:
HInit and HRest need label boundaries, though you could do a flat start using HERest right away.
However data sparsity could make some of your models not as robust as they are when HInit/HRest is used.
If you have enough data, you can try.
Javi.
2009/8/8 Sébastien Le Maguer
<Sebastien.Le_Maguer@xxxxxxxx>
Hello,
I don't think so because how can you extract the part, corresponding to the phoneme, in the signal without it ?
Regards,
Sebastien
On 09/08/2009 00:02, Stas wrote:
Hello,
What is an influence of HInit in HTS-demo_CMU-ARCTIC-SLT/scripts/Training.pl (**) script on the final quality
of models, and as a result, on the quality of the generated speech?
(**) I mean this piece of code in Training.pl
#######################################
# HInit & HRest (initialization & reestimation)
if ($IN_RE) {
...
}
########################################
In other words, can I use row data described by corresponding labels without start/end times of phonemes, but rather using just
phonetic transcriptions of data?
------instead of this------- --just this-
start_time end_time ah ---> ah
Thank You and Best Regards,
Stas
--
Xavi.