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

[hts-users:03683] Re:


Hello,
                  After system building in our language for same hour of durations with EM and DAEM algorithm we observed that size of DAEM is less compared to size of EM.

Can you please tell me is there any specific reason of it?


On Tue, Feb 19, 2013 at 7:04 AM, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx> wrote:
Hi,

It's depend on database.

Regards,
Keiichiro Oura

2013/2/19 Nirmesh Shah <supernirmesh@xxxxxxxxx>:
> Hello,
> So is it 10 is enough? or more you should prefer?
> Regards,
> Nirmesh Shah
>
>
> On Mon, Feb 18, 2013 at 8:07 PM, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
> wrote:
>>
>> Hi,
>>
>> > So I change DAEMNITER=5 So is it going to affect DAEM algorithm(due to
>> > less
>> > no of iteration)?
>>
>> Number of iteration affect the quality of the synthesized speech.
>>
>> > and this DAEM algorithm, Does it work for log F0 re-estimation also in
>> > system?
>>
>> Yes.
>> Spectrum (mel-cep), pitch (log F0), and duration are estimated in
>> HERest simultaneously.
>>
>> Regards,
>> Keiichiro Oura
>>
>>
>> 2013/2/18 Nirmesh Shah <supernirmesh@xxxxxxxxx>:
>> > Dear Sir,
>> >          While making monophone model with the use of DAEM algorithm, It
>> > is
>> > taking so much time because DAEMNITER=10 and NITER=5, So at the
>> > execution of
>> > HERest re-estimation is done for 50 times as it is having structure like
>> > this,
>> >
>> >  if ($daem) {
>> >       for ( $i = 1 ; $i <= $daem_nIte ; $i++ ) {
>> >          for ( $j = 1 ; $j <= $nIte ; $j++ ) {
>> >
>> >             # embedded reestimation
>> >             $k = $j + ( $i - 1 ) * $nIte;
>> >             print("\n\nIteration $k of Embedded Re-estimation\n");
>> >             $k = ( $i / $daem_nIte )**$daem_alpha;
>> >             shell("$HERest{'mon'} -k $k -H $monommf{'cmp'} -N
>> > $monommf{'dur'} -M $model{'cmp'} -R $model{'dur'} $lst{'mon'}
>> > $lst{'mon'}");
>> >          }
>> >       }
>> >    }
>> > So I change DAEMNITER=5 So is it going to affect DAEM algorithm(due to
>> > less
>> > no of iteration)?
>> > and this DAEM algorithm, Does it work for log F0 re-estimation also in
>> > system?
>> > Thanks in advance
>> >
>> > Regards,
>> > Nirmesh Shah
>> >
>> > On Fri, Jan 11, 2013 at 4:50 PM, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx>
>> > wrote:
>> >>
>> >> Hi,
>> >>
>> >> Yes.
>> >> Default setting is EM algorithm.
>> >>
>> >> The other options can be shown by
>> >>
>> >> ./configure --help
>> >>
>> >> Regards,
>> >> Keiichiro Oura
>> >>
>> >>
>> >> 2013/1/11 Nirmesh Shah <supernirmesh@xxxxxxxxx>:
>> >> > Thanks,
>> >> > If I don't write it then does it mean that EM algorithm is used?
>> >> >
>> >> >
>> >> > On Fri, Jan 11, 2013 at 2:04 PM, Keiichiro Oura
>> >> > <uratec@xxxxxxxxxxxxxxx>
>> >> > wrote:
>> >> >>
>> >> >> Hi,
>> >> >>
>> >> >> You can try DAEM training in the HTS demo script by following
>> >> >> command
>> >> >> line.
>> >> >> ./configure DAEM=1
>> >> >>
>> >> >> Regards,
>> >> >> Keiichiro Oura
>> >> >>
>> >> >>
>> >> >> 2013/1/11 Nirmesh Shah <supernirmesh@xxxxxxxxx>:
>> >> >> > Hello,
>> >> >> >     I want to know how to use DAEM algorithm in hts,as I have seen
>> >> >> > on
>> >> >> > hts
>> >> >> > website that HERest command support DAEM, does it mean whenever I
>> >> >> > use
>> >> >> > that
>> >> >> > command ,Is it going to use DAEM algorithm for re-estimation? and
>> >> >> > what
>> >> >> > if I
>> >> >> > don' want to use DAEM?
>> >> >> >
>> >> >> > --
>> >> >> > Thanks & Regards,
>> >> >> > Nirmesh J. Shah
>> >>
>> >> >>
>> >> >
>> >> >
>> >> >
>> >> > --
>> >> > Thanks & Regards,
>> >> > Nirmesh J. Shah
>> >>
>> >>
>> >
>> >
>> >
>> > --
>> > Thanks & Regards,
>> > Nirmesh J. Shah
>> >
>>
>
>
>
> --
> Thanks & Regards,
> Nirmesh J. Shah




--
Thanks & Regards,
Nirmesh J. Shah


Follow-Ups
[hts-users:03684] Re:, Keiichiro Oura
[hts-users:03685] Re:, 那兴宇
References
[hts-users:03566], Nirmesh Shah
[hts-users:03567] Re:, Keiichiro Oura
[hts-users:03568] Re:, Nirmesh Shah
[hts-users:03569] Re:, Keiichiro Oura
[hts-users:03627] Re:, Nirmesh Shah
[hts-users:03628] Re:, Keiichiro Oura
[hts-users:03629] Re:, Nirmesh Shah
[hts-users:03630] Re:, Keiichiro Oura