[mary-users] Problem with voice importer
Ingmar Steiner
ingmar.steiner at inria.fr
Mon Jul 25 09:48:45 CEST 2011
Dear Paul,
On 25.07.2011 00:34, soundmodul at gmx.at wrote:
> Hi Florent!
>
> Thanks!
> You really helped me!
Yes, thank you for the correct solution, Florent! Paul, in case you
would like an explanation, this is due to a bug
(http://mary.opendfki.de/ticket/389) that was discovered after the
release of Mary 4.3. It was fixed right away in the developer code, and
will be included in the next bugfix release (4.3.1?), but right now,
it's not clear if or when that might be released.
In the meantime, it would be possible to use the developer version by
checking out the "4.x-branch" as described at
http://mary.opendfki.de/#SVNaccesstotherepository. Or just fix the lab
files like Florent suggested.
> Additionally, for one of the next tools I had to run your script in
> '/my/path/to/myvoice/phonelab/*.lab' too!
> Now i could run most of the import tools.
>
> Unfortunately, the next problem is the FOCARTTrainer
[...]
> The TimelineReader.java wants to:
> /**
> * Hop the datagrams in the given byte buffer until the one which
> begins at or contains the desired time
> * (time is in samples; the sample rate is assumed to be that of the
> timeline).
>
> Is it possible, that the sample rate of wav files (22050 for me) is the
> problem?
Indeed, you're quite probably correct. Many parts of the Mary code rely
on the assumption that the audio is sampled at 16kHz. This limitation is
unlikely to be fixed in the near future (e.g.
http://mary.opendfki.de/ticket/278), so in the meantime, you will have
to use your favorite audio utility to downsample everything to 16000 Hz.
It would be a good idea to repeat ALL steps of the voicebuilding process
afterwards.
Best wishes,
-Ingmar
>
> Thank you really very much for your help!
> paul
>
>
> Am 2011-07-24 18:35, schrieb fxavier at ircam.fr:
>> Hi,
>>
>> The procedure explain in the VoiceImportTools tutorial is alas not
>> complete. Your error comes because there are commas instead of point in
>> your lab files. So you have to convert those commas into points. In your
>> lab directory, create some sh file and paste this
>>
>>
>> for i in /my/path/to/myvoice/lab/*.lab
>> do
>> echo ${i}
>> sed -i 's/,/./g' /my/path/to/myvoice/lab/*.lab
>> done
>>
>>
>>
>>
>> Just change the path to your voice. The problem is that even the lab files
>> output by the ehmmlabeler (and the CMU lab files as well) comes with
>> commas. This is missing in the VoiceImportTool tutorial but it should be
>> added soon.
>>
>> Hope it helps,
>> Florent
>>
>>
>>
>>
>>
>>
>>> Hi H.Sasse
>>>
>>> Thanks for your reply.
>>> Maybe you did not see my attachment (png) with the java exception:
>>>
>>> #######################################
>>> Computing unit labels for 219 files.
>>> From phonetic label files: /home/ppp/MARY_TTS/MyVoice/lab/*.lab
>>> to unit label files: /home/ppp/MARY_TTS/MyVoice/phonelab/*.lab
>>> Mailformed line found outside of header:
>>> 0,555000 125 _
>>> java.lang.Exception: The component PhoneUnitLabelComputer produced the
>>> following exception:
>>> at
>>> marytts.tools.voiceimport.DatabaseImportMain$8.run(DatabaseImportMain.java:294
>>> Caused by: java.io.IOException
>>> at
>>> marytts.util.data.text.XwavesLabelfileDataSource.parseLabels(XwavesLabelfileDataSource.java:157)
>>> at
>>> marytts.util.data.text.XwavesLabelfileDataSource.<init>(XwavesLabelfileDataSource.java:71)
>>> .
>>> .
>>> .
>>> ########################################
>>>
>>> The "Mailformed line found outside of header" is the second line of
>>> *.lab file.
>>> The first line is a simple # that marks the end of header.
>>>
>>> Hence i see the problem in the XwavesLabelfileDataSource.java.
>>> I would really like to import own voices, but it's not easy.
>>> Thanks, again.
>>>
>>>
>>>
>>> Am 2011-07-24 15:42, schrieb H. Sasse:
>>>> I've done nothing with importing voices, and I'm not familiar with the
>>>> code, but
>>>> I think I may be able to help you get a step further forwards.
>>>>
>>>> On 24/07/2011 13:24, soundmodul at gmx.at wrote:
>>>>> Hi to all!
>>>>>
>>>>> I tried to created my own voice, and so i wrode textfiles (did not
>>>>> find german textfiles for reuse)
>>>> You can find plenty of free texts at Project Gutenburg. I think there
>>>> have been changes
>>>> in how one writes German, so some of them may not be suitable.
>>>> http://www.gutenberg.org/browse/languages/de
>>>>> and recordet the wav files.
>>>>> Than i tried to use the voice import tools.
>>>>> The first tool, that sucks is the PhoneUnitLabelComputer, this tool (I
>>>>> looked into source) searches for strange chars like "^\\s*(......"
>>>> This looks like a regular expression, and it looks like the \ has been
>>>> quoted in accordance
>>>> with
>>>> http://download.oracle.com/javase/7/docs/api/java/util/regex/Pattern.html
>>>> or similar. It means, "look for any whitespace (including none) and
>>>> then collect whatever is
>>>> in the brackets into a variable", so it will skip leading spaces or tabs
>>>> if there are any.
>>>> However, you have not explained why you believe this is the cause of
>>>> your failure.
>>>>> (XwavesLabelfileDataSource.java)
>>>>> But my *.lab files looks like this:
>>>>> ###################
>>>>> #
>>>>> 0,490000 125 _
>>>>> 0,510000 125 d
>>>>> 0,655000 125 aI
>>>>> .
>>>>> .
>>>>> .
>>>>> ###################
>>>>> I wonder, because the *.lab files looks similar than the downloadet
>>>>> example cmu_us.
>>>>> Does anyone know, what I am doing wrong?
>>>> You might get more help if you explain how you tried to use the voice
>>>> import tools.
>>>> What commands, with what arguments, did you use? What results did you
>>>> expect?
>>>> Then what actually happened? The important thing about debugging
>>>> problems like this is being able to reproduce the problem exactly, so it
>>>> can be analysed.
>>>> Saying that something "sucks" doesn't get people much closer to that
>>>> position.
>>>>> Thanks in advance!
>>>> Hugh
>>>> _______________________________________________
>>>> Mary-users mailing list
>>>> Mary-users at dfki.de
>>>> http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users
>>>>
>>> _______________________________________________
>>> Mary-users mailing list
>>> Mary-users at dfki.de
>>> http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users
>>>
>>>
>> _______________________________________________
>> Mary-users mailing list
>> Mary-users at dfki.de
>> http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users
>>
>
> _______________________________________________
> Mary-users mailing list
> Mary-users at dfki.de
> http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users
--
Ingmar Steiner
Postdoctoral Researcher
LORIA Speech Group, Nancy, France
National Institute for Research in
Computer Science and Control (INRIA)
More information about the Mary-users
mailing list