[hts-users:02961] Re: Inconsistency between final time label and total # of obs in cmp file

[Ninh Khanh Duy <ninhkhanhduy@xxxxxxxxx>]

Hi,

Thank you for your explanation and suggestion.

Regards,

Ninh K.D.

On Wed, Jul 20, 2011 at 12:15 PM, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx> wrote:

Hi,

The length of utterances in utt files and waveforms are sometimes
mismatched in CMU ARCTIC DATABASE.
But, It seems that it affects little impact to the performance of the system.
For example, although 23 frames in a0005 are not used for only
initialization step of ``pau'' (HInit), these frames are used in
following step (HRest, HERest, etc.).
If you can't ignore them, please use DAEM option in 2.2.

Regards,
Keiichiro Oura


2011/7/20 Ninh Khanh Duy <ninhkhanhduy@xxxxxxxxx>:

> Hi,
> I use speaker-dependent HTS-demo_CMU-ARCTIC-SLT version 2.1.1. To name just
> a few label files in /data/labels/full folder have largely wrong number of
> frames as follows:
> cmu_us_arctic_slt_a0005.lab: shorter than cmp file 23 frames
> cmu_us_arctic_slt_a0008.lab: ~ 23 frames
> cmu_us_arctic_slt_a0011.lab: ~ 18 fr
> cmu_us_arctic_slt_a0026.lab: ~ 34 fr
> ...
> and many others.
> BR,
> Ninh K.D.
>
> On Wed, Jul 20, 2011 at 12:15 AM, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
> wrote:
>>
>> Hi,
>>
>> Which label has wrong number of frames?
>> Please, tell me label dir/file name and demo script name.
>>
>> Regards,
>> Keiichiro Oura
>>
>>
>> 2011/7/19 Ninh Khanh Duy <ninhkhanhduy@xxxxxxxxx>:
>> > Hi,
>> > I used HTS demo 2.1.1 and found that there are several cases the final
>> > time
>> > label in label file is much shorter/longer than the product of the total
>> > #
>> > of observations in cmp file and the frame rate. Why does this
>> > inconsistency
>> > exist? Does this affect the performance of the system?
>> > Regards,
>> > Ninh K.D.
>>
>
>