[hts-users:02220] Re: problem using hts voice in festival

Hello,

Thank you for your fast response.

I have tried your second solution to the problem…

Using the files from the cmu_us_slt_arctic_hts voice

And converting with the swab only the .pdf files (dur, lf0 mgc)

I am still getting the same error:

Segmentation fault (core dumped)

And a file with the name festival.exe.stackdump has been created in the main directory with the following content:

Exception: STATUS_ACCESS_VIOLATION at eip=004BB196

eax=00000007 ebx=00000000 ecx=00000002 edx=00000005 esi=00F3E5C0 edi=3FD163BF

ebp=0022BDC8 esp=0022BDA0 program=C:\cygwin\home\festival\festival\src\main\festival.exe, pid 11556, thread main

cs=001B ds=0023 es=0023 fs=003B gs=0000 ss=0023

Stack trace:

Frame Function Args

0022BDC8 004BB196 (00F3E0F0, 0022C670, 00000000, 00000000)

0022C278 00451A8C (00D81684, 00D6D624, 00D815B4, 00D8161C)

0022C6B8 00452587 (7E9D3068, 7EA13FE8, 7EA13E38, 00000000)

0022C708 00578294 (7E96F6F8, 7EA13FE8, 7EA13D68, 0022C784)

0022C728 005D4E06 (0022C780, 0022C784, 7E9D42F8, 00000000)

0022C778 00578415 (7EA13E58, 7EA13FE8, 7E9D44A8, 0022C7F4)

0022C798 005D4E06 (0022C7F0, 0022C7F4, 7E9D3028, 7E9D2FA8)

0022C7E8 00578415 (7E7F1758, 7EA11F98, 00230000, 0022C864)

0022C808 005D4E06 (0022C860, 0022C864, 0022C828, 0057CFF2)

0022C858 00578415 (7E7F0718, 7E9D3098, 0022C890, 0022C8D4)

0022C878 005D4E06 (0022C8D0, 0022C8D4, 0022C8A8, 611010E8)

0022C8C8 00578415 (7E7F0B78, 7E9D2F48, 00000000, 3FE91111)

0022C948 0057A170 (00000000, 00574310, 0022C968, 0057CDC0)

0022C988 0057A4F7 (00000001, 00000000, 00000000, 00404975)

0022C9A8 00588F19 (00000001, 0022CAB0, 00000001, 0022CCB0)

0022CCE8 00402CD4 (00000001, 611696B0, 00B10090, 00000000)

End of stack trace (more stack frames may be present)

Is there anything else I should try in order to fix this error or do I have to use your first solution with the hts_engine and recompiling festival (make)???

Thank you in advance,

Alexandros Lazaridis.

From: Sergio Oller [mailto:sergioller@xxxxxxxxx]
Sent: Wednesday, September 30, 2009 1:09 PM
To: hts-users@xxxxxxxxxxxxxxx
Subject: [hts-users:02219] Re: problem using hts voice in festival

Hi,

I have adapted the hts_engine 1.01 to work as a festival module. You can download it from here:

http://www.talp.cat/festcat/hts_engine_new/

I sent it to Festival developpers, so I hope it will replace the current module soon (if they like it).

So with this new hts_engine module there is no need to swab anything and .win1 .win2 .win3 files work. Check that webpage to download a demo voice also.

===

If you don't want to use this new hts_engine, you can convert the HTS demo to a festival voice by doing this:

Window files:
Don't use the .win1 .win2 and .win3 files. Use instead mcep_acc.win mcep_dyn.win lf0_acc.win and lf0_dyn.win from an old demo voice (available here, for example):
Why?
mcep_acc.win &Company files have float numbers stored in binary format.
.win2 and .win3 files have float numbers in ascii format.
.win1 files have an extra number which is not specified in older hts voices.

The current festival hts_engine expects window files to be in float binary format, not ASCII.

inf and pdf:
inf files have text information, don't swab them.
You have to swab .pdf files only (all of them).

I think I have not forgotten anything. I hope I have been able to help you.
If you have any doubt or problem, please ask.

Sergio Oller
Student working at UPC

On Wed, Sep 30, 2009 at 10:27, Alexandros Lazaridis <alaza@xxxxxxxxxxxxxxxxx> wrote:

Hello,

I am trying to use the files from HTS-demo_CMU-ARCTIC-SLT into the voice of festival cmu_us_slt_arctic_hts

I have done the following according to this message

http://hts.sp.nitech.ac.jp/hts-users/spool/2009/msg00044.html

swab +f /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/mgc.pdf > /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/mgcs.pdf

swab +f /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/lf0.pdf > /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/lf0s.pdf

swab +f /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/tree-dur.inf > /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/trees-dur.inf

swab +f /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/tree-lf0.inf > /cygdrive/c/htk/HTS-demo_CMU-ARCTIC-SLT_olokliromeno/voices/qst001/ver1/trees-lf0.inf

from hts demo -------à to festival

dur.pdf                 --> duration.pdf

lf0s.pdf                --> lf0.pdf

mgcs.pdf             --> mcep.pdf

label.feats           --> feat.list

trees-dur.inf        --> trees-dur.inf

trees-lf0.inf         --> trees-lf0.inf

tree-mgc              --> trees-mcep.inf

lf0.win1                --> lf0_acc.win

lf0.win2                --> lf0_dyn.win

lf0.win3 (add)

mgc.win1              --> mcep_acc.win

mgc.win2              --> mcep_dyn.win

mgc.win3 (add)

Now I am getting the following error when I am trying to synthesize in festival

$ ./festival

Festival Speech Synthesis System 1.96:beta July 2004

For details type `(festival_warranty)'

festival> (voice_cmu_us_slt_arctic_hts)

cmu_us_slt_arctic_hts

festival> (SayText "hello, have a nice day")

10 [main] festival 10992 _cygtls::handle_exceptions: Error while dumping state (probably corrupted stack)

Segmentation fault (core dumped)

And exits festival.

I have read in a message in festival mailing list (http://lists.berlios.de/pipermail/festlang-talk/2009-April/003004.html) about x2x conversion…

Is there something I am not doing right?

Thank you in advance

Alexandros Lazaridis.