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

[hts-users:01973] Re: forced alignment with htk

Hi,

In HTS-2.1, forced alignment of HSMM is not supported.
You should use HMM instead of HSMM.

Regards,
Keiichiro Oura

meron wrote:

Hi,

You can try
HVite -T 1 -C trn.cfg -f -H model.mmf -S file_list -l out_lab_dir -y lab -L in_lab_dir dict full_list 

where:
full_list is the list of all the context dependent models, and dict is similar, except that each line has the model name twice - you can create it by executing: 
paste full_list full_list > dict
Note that this will NOT automatically add silences between "words" - silences need to be explicitly listed in your input lab file. 

Cheers,
Joram




--- On *Thu, 5/7/09, Cenk Demiroglu /<cenk_demiroglu@xxxxxxxxxxx>/* wrote:

    From: Cenk Demiroglu <cenk_demiroglu@xxxxxxxxxxx>
    Subject: [hts-users:01969] forced alignment with htk
    To: hts-users@xxxxxxxxxxxxxxx
    Date: Thursday, May 7, 2009, 12:15 PM


    I'm trying to do forced alignment using the fullcontext models generated
    after the hts training process. Although I can force align with the
    monophone
    models, I cannot force align with the context dependent phones. Has
    anyone
    tried that before? Does anyone have the commandline for forced-alignment
    after training with hts a lab file and a cmp file from the training
    data are used?

    Thanks,
    Cenk

    ------------------------------------------------------------------------
    Insert movie times and more without leaving Hotmail®. See how.
    <http://windowslive.com/Tutorial/Hotmail/QuickAdd?ocid=TXT_TAGLM_WL_HM_Tutorial_QuickAdd1_052009>
References
[hts-users:01970] Re: forced alignment with htk, meron