<html><head></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space; ">Hello Marcela,<div><br></div><div>Thank you for your quick response and sorry for coming back so late. I was busy on some other projects and I wanted to make sure I could solve all the problems before coming back to you and the user list. I will share here some of the troubles we experienced and the solutions we found.</div><div><br></div><div>The "styleDefinition.txt" filename must be included in the "AllophoneExtractor" with the complete path!</div><div>The code in "AllophonesExtractor.java" suggest that the default path of the styleDefinition.txt file is in the "mary<i>" </i>folder rather than the main voice building directory (CONFIGDIR variable defined in "DatabaseLayout.java"). Anyway this part of the code managing the "styleDefinition.txt" file was inactivated (commented) in the original code (Mary 5.0), making the file complete path in the "AllophonesExtractor" settings required.</div><div><br></div><div>Also, after editing the settings, the "VoiceImport" module needs to be restarted, since it takes the "styleDefinition.txt" file into account only at startup.</div><div><br></div><div>Later during the training the "PhoneUnitFeatureComputer" failed, because the label included in the "prompt_allophones" XML files was refused. We realized that the "style" label wasn't included in the MaryXML Schema Definition file ($MARYBASE/marytts-common/src/main/resources/marytts/util/dom/MaryXML.xsd).</div><div>We then added the line:</div><div>"<xsd:attribute name="style" type="xsd:string"/>"</div><div>in the "prosody" element definition to solve the problem.</div><div><br></div><div>The further steps could then well be processed until the end of the HMM training process. Although we weren't yet able to verify that the style was well taken into account during the training.</div><div><br></div><div>Thanks again for your help!</div><div>Best regards,</div><div>Marc</div><div><br></div><div><br><div><div>On Feb 6, 2013, at 15:39 , Marcela Charfuelan wrote:</div><br class="Apple-interchange-newline"><blockquote type="cite"><div>Dear Marc Evrard,<br><br>The names of your files in the text directory should reflect what the styles file indicate, for example if you have files like:<br>file_mydb_aaa_001.txt<br>file_mydb_aaa_002.txt<br>file_mydb_aaa_003.txt<br>...<br>file_mydb_bbb_001.txt<br>file_mydb_bbb_002.txt<br>file_mydb_bbb_003.txt<br>...<br><br>then the styleDefinition.txt file in your voice building directory (not in the text dir) should contain (supposing that aaa correspond to angry and bbb to happy):<br><br>file_mydb_aaa_* = angry<br>file_mydb_bbb_* = happy<br><br><br>The style feature has been used in building unit selection voices, but so far we have not used in the building of HMM voices.<br><br>Regards,<br>Marcela.<br><br><br>On 02/04/2013 05:46 PM, Marc Evrard wrote:<br><blockquote type="cite">Hello,<br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">I'm currently trying to build a voice in MaryTTS using 2 almost identical corpora recorded with different style attributes.<br></blockquote><blockquote type="cite">The building of the individual corpora as different voices without any style attribute is working fine, but I would like to build a single voice using these 2 corpora.<br></blockquote><blockquote type="cite">I couldn't find any info in the wiki (dfki and Github), nor in the mary-users mailing lists about this procedure.<br></blockquote><blockquote type="cite">In the java source code of the AllophoneExtractor, I saw that a ".styleDefinitionFile" file must be include somewhere. I assumed it needs to be included in the "text" folder using the same name as the corresponding sentence, but doing so doesn't change the XML output file in the "prompt_allophones" folder.<br></blockquote><blockquote type="cite">Could someone give me a hint, or a link to some documentations describing this process?<br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite">Thanks in advance!<br></blockquote><blockquote type="cite">Best regards,<br></blockquote><blockquote type="cite">Marc<br></blockquote><blockquote type="cite">_______________________________________________<br></blockquote><blockquote type="cite">Mary-users mailing list<br></blockquote><blockquote type="cite"><a href="mailto:Mary-users@dfki.de">Mary-users@dfki.de</a><br></blockquote><blockquote type="cite"><a href="http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users">http://www.dfki.de/mailman/cgi-bin/listinfo/mary-users</a><br></blockquote><blockquote type="cite"><br></blockquote><blockquote type="cite"><br></blockquote><br><br>-- <br>_______________________________________________________________<br> Marcela Charfuelan, Researcher, DFKI GmbH<br> Projektbuero Berlin, Alt-Moabit 91c, D-10559 Berlin, Germany<br> Phone: +49 (0)30 23895-1821<br> URL :<a href="http://www.dfki.de/~charfuel/">http://www.dfki.de/~charfuel/</a><br>_______________________________________________________________<br> Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH<br> Firmensitz: Trippstadter Strasse 122, D-67663 Kaiserslautern<br> Geschaeftsfuehrung:<br> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)<br> Dr. Walter Olthoff<br> Vorsitzender des Aufsichtsrats:<br> Prof. Dr. h.c. Hans A. Aukes<br> Amtsgericht Kaiserslautern, HRB 2313<br>_______________________________________________________________<br><br></div></blockquote></div><br></div></body></html>