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

[hts-users:02082] Re: CMLLR


I'm afraid I don't know.
Did send an e-mail to HTK mailing list about this?

Regards,
Junichi

On 10 Jul 2009, at 12:47, pollyvin@xxxxxxxxx wrote:

Thanks for your answers Dr. Junichi.

I guess there is no straightforward method available/foreseen in HTK/ HTS for a multiple regression class based linear transform?

Cheers,
-p.

On Fri, Jul 10, 2009 at 11:23 AM, Junichi Yamagishi <jyamagis@xxxxxxxxxxxx > wrote:
Hi,

If the CMLLR transform is global, there are two ways.

A) Use of InputXForm
you need to change text format of the transform
from AdaptXForm to InputXForm manually.
See HTKBook for the format.
Then specify the InputXForm in your config like this
 HPARM: MATTRANFN = /home/test/lintran.mat
and apply HCopy to cmp files for transforming features.

B) Use of SideXForm
You may directly specify the AdaptXForm as sideXForm
via SIDEXFORMMASK and SIDEXFORMEXT in your config
and apply HCopy to cmp files for transforming features.

Regards,
Junichi
CSTR




On 9 Jul 2009, at 11:46, pollyvin@xxxxxxxxx wrote:

Hi,

I would like to evaluate the feature space adaptation techniques of hts. Given a trained linear transform and hmm model, what would be your advice on how to write out the transformed features of a training utterance with hts?

Thanks,
-p.


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





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


References
[hts-users:02079] CMLLR, pollyvin
[hts-users:02080] Re: CMLLR, Junichi Yamagishi
[hts-users:02081] Re: CMLLR, pollyvin