[hts-users:00491] Re: HTK 2.0 hts_engine problem

["Nicholas Volk" <nvolk@xxxxxxxxxx>]

Hi Heiga,

When I comment the byte swapping line in misc.cpp
I get the problem I described in my earlier mail:

   //HTS_ByteSwap(p, size, block);

If the line is uncommented I get an error:
"Error: LoadModelFiles: #duration pdf must be positive value (-419364864)."

I think that the error is caused by endianism.
Thus I'd expect the commented version work fine, but the content
of the created raw is not really speech.

My windows should be ok:

$ for i in `ls *win?`; do echo $i; cat $i; done
lf0.win1
1 1.0
lf0.win2
3 -0.5 0.0 0.5
lf0.win3
3 1.0 -2.0 1.0
mcp.win1
1 1.0
mcp.win2
3 -0.5 0.0 0.5
mcp.win3
3 1.0 -2.0 1.0


And a slightly different issue:
There's a bug in the Festival HTS version of RMS voice (at least):

$ festival                                                                
     Festival Speech Synthesis System 1.96:beta July 2004                 
          Copyright (C) University of Edinburgh, 1996-2004. All rights
reserved.          For details type `(festival_warranty)'
festival> (voice_nitech_us_rms_arctic_hts )
SIOD ERROR: unbound variable : ::
closing a file left open:
/home/Nicholas/festival/lib/voices/us/nitech_us_rms_arctic_hts/festvox/nitech_us_rms_arctic_hts.scm

(There's a mistyped comment that ought to be ";;".)

br,
  Nicholas

> Hi,
>
> Nicholas Volk wrote:
>
>> I'm able to build new voices (pdf and inf files) with hts 2.0 but I'm
>> not
>> able to produce speech files with hts_engine. I have built a new voice
>> with 2.0 that works with older hts_engines (v. 1.1.1 or so). Similar
>> problems exists with the downloadable voices. I loaded the RMS voice and
>> tried tpo use it with:
>>
>> $ cd ~/festival/lib/voices/us/nitech_us_rms_arctic_hts/hts
>> $ ~/htk-3.4/hts_engine/hts_engine -td trees-dur.inf -tf trees-lf0.inf
>> -tm
>> trees-mcep.inf -md duration.pdf -mm mcep.pdf -mf lf0.pdf -df lf0.win1
>> -df
>> lf0.win2 -df lf0.win3 -dm mcp.win1 -dm mcp.win2 -dm mcp.win3 -or foo.raw
>> -ot trace.txt alice01.lab
>>
>> (lab, inf and pdf files taken from SLT example)
>>
>> The raw file is successfully created but the content is not what I'd
>> expect. Can some produce speech with the above commands (or similar) or
>> does someone have an idea what's wrong with them?
>
> Downloadable voices are in little endian because Festival 1.95 & 1.96 are
> based on old hts_engine (1.1).
> In the current version (~/htk-3.4/hts_engine/hts_engine) model files (.pdf
> files) must be in big endian.
>
> It should be noted that format of window coefficients files is changed.
> In the previous versions window files are in binary but in the current
> version window files should be in ascii.
> Please check endian and window file format of your files.
>
> Regards,
>
> Heiga ZEN (Byung Ha CHUN)
>
> --
> ------------------------------------------------
>  Heiga ZEN     (in Japanese pronunciation)
>  Byung Ha CHUN (in Korean pronunciation)
>
>  Department of Computer Science and Engineering
>  Nagoya Institute of Technology
>  Gokiso-cho, Showa-ku, Nagoya 466-8555 Japan
>
>  http://kt-lab.ics.nitech.ac.jp/~zen
> ------------------------------------------------
>