[mary-users] Creating Voice for/with Mary - stuck on EHMM-labelling

[Margot.Mieskes at gmx.net]

Dear all,

my name is Margot and I'm new to this mailing list. I'm sorry, if I am repeating a question, but I could not find a solution to my problem. Maybe someone can help me.

I'm trying to build a new voice using the Mary VoiceImport-GUI. I successfully ran

* PraatPitchmarker
* MCEPMaker
* AllophonesExtractor

When I start the "EHMMLabeler"-Step it prints to the console all the commands that it apparently performs and also ends with "Label file Generation Successfully completed using EHMM !", but in reality the ehmm/lab/-folder is empty, which means that subsequent steps also fail - run on 0 files. 

I checked all the steps that are performed during the EHMMLabeler step and performed them manually. This works fine, until I reach the baum-welch re-estimation step, which calls the "ehmm" script (MARYDIR/lib/external/ehmm/bin/ehmm). This runs for a while and its output is:

Sequential Flag set to 1
Read Flag set to 0
Feature Path: /path/to/VoiceImportDir/ehmm/feat
Feature exten: ft
Fully-Connected Flag set to 0
Perturbation  Flag set to 0
Skip First State Flag set to 0
SHORT PAUSE WOULD BE INTRODUCED WHEN diff is < 0.01
NoWords: 51
TotalSt: 37144
FDim: 13
FOR STATE: 34104 RESETTING THE CONNECTIONS TO 2
Total Prompts: 144
Processing /path/to/VoiceImportDir/ehmm/feat/rec_001.ft
Feat file loading.....
In emissions
Killed.

But other than that, nothing happens. So I'm basically stuck again. Did anyone experience this? Can anyone help me with this? 
I'm running Mary 4.1.1 on Ubuntu. If you need any other information, please let me know.

Greetings and thanks in advance,
Margot.



-- 
GMX DSL SOMMER-SPECIAL: Surf & Phone Flat 16.000 für nur 19,99 Euro/mtl.!*
http://portal.gmx.net/de/go/dsl