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

[hts-users:02549] Re: Utterance size problem


Thanks Keiichiro,

But i have two question about this case.
1. how much Ram does need? my biggest wave file is about 2.5MB. can 16GB be enough or more need?
2. What is the effect of this part on synthesizing? Does ``HSMM-based Forced-alignment using WFST'' affect on quality of synthesis? if your answer is yes can you introduce any article about it?  

Best Regards,
-Ali

On Fri, Jul 9, 2010 at 11:37 AM, Keiichiro Oura <uratec@xxxxxxxxxxxxxxx> wrote:
Hi,

You should set ``No-silent GV flag'' to off by typing ``configure NOSILGV=0''.
A huge RAM is required for ``HSMM-based Forced-alignment using WFST''.

Regards,
Keiichiro Oura

2010/7/7 ali azimizadeh <ali.azimizadeh@xxxxxxxxx>:
> Hi,
> I changed the HTS_Demo for HTS 2.1.1 to make a hts voice in Persian with my
> dataset but there is a little problem. my system properties are:
> CPU: Pentium 4 Dual core 2.26GHz
> Ram: 4G
> Swap: 15G
> OS: CentOS 5.3
> Most of my wav file size are big. its range is from 33k to 2M. the main
> problem occurs in "forced alignment using WFST for no-silent GV".
> when the raw file is less than 300k everything is OK. but when it is bigger
> than 300k, at first the memory is fulled and then swap goes to maximum size
> and CPU doesn't work more.
> Often, $gvfaldir/$base.tmp1.sym and $gvfaldir/$base.tmp2.fst are made by
> fstcompose.
> note that i can't spite the utterance files into shorter ones more than
> this. do you have any hint?
> and also, do you think that if the Ram is upgraded ,for example, to 8G the
> problem is solved?
>
> Regards,
> -Ali
>
>
>



Follow-Ups
[hts-users:02550] Re: Utterance size problem, Simon King
References
[hts-users:02538] Utterance size problem, ali azimizadeh
[hts-users:02547] Re: Utterance size problem, Keiichiro Oura