[hts-users:02415] Re: OpenFST problem

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

Hi,

Sorry for the delayed answer. The file sizes are:

nvolk@server:~/HTS-demo_BL_CY_OW2> ls -l `find . -name "*ts_190*.fst"`
-rw-rw-r-- 1 nvolk nvolk 115844638 Mar  9 11:50
./gv/qst001/ver1/fal/bl_cy_ow_clunits_190.tmp2.fst
-rw-rw-r-- 1 nvolk nvolk   1376882 Mar  9 11:50
./gv/qst001/ver1/fal/bl_cy_ow_clunits_190.tmp3.fst
-rw-rw-r-- 1 nvolk nvolk 276488046 Mar  9 10:04
./models/qst001/ver1/cmp/HFst/bl_cy_ow_clunits_190.fst
-rw-rw-r-- 1 nvolk nvolk   2434575 Mar  9 10:04
./models/qst001/ver1/dur/HFst/bl_cy_ow_clunits_190.fst
nvolk@server:~/HTS-demo_BL_CY_OW2>

br,
  Nicholas

> Hi Nicholas,
>
> How big is your bl_foo_ow_clunits_190.fst file?
>
> In my data, these fst files sometimes reach 1 to 2GBytes and
> thus I needed to use 8 or 10GBytes for running WFST..
>
> One possible choice is trim silences of the audio data more and
> to make the data more compact.
>
> Regards,
> Junichi
>
>
>
> On 10 Feb 2010, at 06:12, Nicholas Volk wrote:
>
>> Hi,
>>
>> The machine is 64-bit and I haven't defined -m32 anywhere either. So I
>> assume it is 64-bit.
>>
>> $ uname -a
>> Linux somehost.bitlips.fi 2.6.18-92.1.10.el5 #1 SMP Tue Aug 5
>> 07:42:41 EDT
>> 2008 x86_64 x86_64 x86_64 GNU/Linux
>>
>> $ gcc --version
>> gcc (GCC) 4.1.2 20080704 (Red Hat 4.1.2-46)
>>
>> $  g++ --version
>> g++ (GCC) 4.1.2 20080704 (Red Hat 4.1.2-46)
>>
>> Also the openfst's ./configure seems refers to 64 bits with lines
>> such as:
>> checking whether the g++ linker (/usr/bin/ld -m elf_x86_64) supports
>> shared libraries... yes
>>
>> I managed to reproduce the error with another voice using large audio
>> files, so it seems to be a file size issue. Also build succeeds when
>> gv is
>> turned off.
>>
>> br,
>>  Nicholas
>>
>>> Hi,
>>>
>>> Could you try to compile OpenFst-1.1 with 64 bit?
>>>
>>> Regards,
>>> Keiichiro Oura
>>>
>>> Nicholas Volk wrote:
>>>> Hi,
>>>>
>>>> I'm getting the following error when doing the globabl variance
>>>> with HTS
>>>> 2.1.1 beta:
>>>>
>>>> ...
>>>> Forsed alignment from bl_foo_ow_clunits_170.fst using
>>>> OpenFst...done.
>>>> Forsed alignment from bl_foo_ow_clunits_180.fst using
>>>> OpenFst...done.
>>>> Forsed alignment from bl_foo_ow_clunits_190.fst using
>>>> OpenFst...Error in
>>>> $HOME/openfst-1.1/bin/fstcompose
>>>> /home/nvolk/HTS-demo_BL_FOO_OW2/gv/qst001/ver1/fal/
>>>> bl_foo_ow_clunits_190.tmp2.fst
>>>> /home/nvolk/HTS-demo_BL_FOO_OW2/gv/qst001/ver1/fal/
>>>> bl_foo_ow_clunits_190.tmp3.fst
>>>> /home/nvolk/HTS-demo_BL_FOO_OW2/gv/qst001/ver1/fal/
>>>> bl_foo_ow_clunits_190.tmp4.fst
>>>>
>>>> Not sure why this happens. Nothing seemingly wrong with the audio.
>>>> The
>>>> audio file in question is way longer than any of the previous files
>>>> (around 30s of speech with sample rate 16k). Is there a chance that
>>>> fstcompose chokes on big input files (runs out of memory etc)?
>>>>
>>>> br,
>>>>  Nicholas
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> __________ Information from ESET NOD32 Antivirus, version of virus
>>>> signature database 4842 (20100206) __________
>>>>
>>>> The message was checked by ESET NOD32 Antivirus.
>>>>
>>>> http://www.eset.com
>>>>
>>>>
>>>
>>>
>>>
>>> __________ Information from ESET NOD32 Antivirus, version of virus
>>> signature database 4842 (20100206) __________
>>>
>>> The message was checked by ESET NOD32 Antivirus.
>>>
>>> http://www.eset.com
>>>
>>>
>>>
>>>
>>
>>
>>
>>
>
>
> --
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.
>
>
>