[hts-users:03671] voice building error

[Andrei Barbos <andreibarbos.banc@xxxxxxxxx>]

Hi everybody,

I am using HTS2.2 and I want to build a voice for Romanian but I just can't complete the voice building process because of an error issued by HHEd. The strange thing is that it just stops working without any error message as seen in the following excerpt from the log file.

--------------------------------------------------
Clustering state 2 lsf
--------------------------------------------------

/home/andrei/Apps/CSTRVoiceClone/bin/HHEd -A -B -C /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//config/general.conf -C /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//config/clust.conf -D -V -H /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/EmbeddedTraining/2/full.cmp.mmf.lsf -T 1 -i -m -a 5.00000000000000000000 -p -r 1 -s -w /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/ContextClustering-lsf/2/clustered.cmp.mmf /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/ContextClustering-lsf/cluster.lsf.hed.2 /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//list/full.lst 

HTK Configuration Parameters[9]
 Module/Tool Parameter Value
# SHRINKOCCTHRESH Vector 6 1488.0 1100.0 600.0 2000.0 2000.0 2000.0
# MINLEAFOCC 5
 # MAXSTDDEVCOEF 10
# DURVARFLOORPERCENTILE 1.000000
# APPLYDURVARFLOOR TRUE

# VFLOORSCALESTR Vector 6 0.01 0.01 0.01 0.01 0.01 0.01
 # APPLYVFLOOR TRUE
# NATURALWRITEORDER TRUE
# NATURALREADORDER TRUE


HTK Version Information
Module Version Who Date : CVS Info
 HHEd 3.4.1 CUED 12/03/09 : $Id: HHEd.c,v 1.105 2011/06/16 05:18:47 uratec Exp $

HShell 3.4.1 CUED 12/03/09 : $Id: HShell.c,v 1.15 2011/06/16 04:18:29 uratec Exp $
HMem 3.4.1 CUED 12/03/09 : $Id: HMem.c,v 1.11 2011/06/16 04:18:29 uratec Exp $
 HLabel 3.4.1 CUED 12/03/09 : $Id: HLabel.c,v 1.12 2011/06/16 04:18:29 uratec Exp $

HMath 3.4.1 CUED 12/03/09 : $Id: HMath.c,v 1.15 2011/06/16 04:18:29 uratec Exp $
HSigP 3.4.1 CUED 12/03/09 : $Id: HSigP.c,v 1.8 2011/06/16 04:18:29 uratec Exp $
 HWave 3.4.1 CUED 12/03/09 : $Id: HWave.c,v 1.11 2011/06/16 04:18:29 uratec Exp $

HAudio 3.4.1 CUED 12/03/09 : $Id: HAudio.c,v 1.9 2011/06/16 04:18:28 uratec Exp $
HVQ 3.4.1 CUED 12/03/09 : $Id: HVQ.c,v 1.8 2011/06/16 04:18:29 uratec Exp $
 HModel 3.4.1 CUED 12/03/09 : $Id: HModel.c,v 1.42 2011/06/16 05:07:56 bonanza Exp $

HParm 3.4.1 CUED 12/03/09 : $Id: HParm.c,v 1.15 2011/06/16 04:18:29 uratec Exp $
HUtil 3.4.1 CUED 12/03/09 : $Id: HUtil.c,v 1.31 2011/06/16 04:18:29 uratec Exp $
 HAdapt 3.4.1 CUED 12/03/09 : $Id: HAdapt.c,v 1.64 2011/06/16 04:15:50 uratec Exp $

HHEd
 12915/12915 Models Loaded [7 states max, 2 mixes max]

RO 0.00 ''
 Setting outlier threshold for clustering
 RO->LS /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/EmbeddedTraining/cmp.sts

 and loading state occupation stats
 Stats loaded for 12916 models
 Mean Occupation Count = 4.260795

TR 1
 Adjusting trace level
--------------------------------------------------
 Realtime 0:00:01

--------------------------------------------------
Error in this command : /home/andrei/Apps/CSTRVoiceClone/bin/HHEd -A -B -C /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//config/general.conf -C /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//config/clust.conf -D -V -H /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/EmbeddedTraining/2/full.cmp.mmf.lsf -T 1 -i -m -a 5.00000000000000000000 -p -r 1 -s -w /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/ContextClustering-lsf/2/clustered.cmp.mmf /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//training/ContextClustering-lsf/cluster.lsf.hed.2 /home/andrei/AppData/nsr_data/Spannish-v1/test_voice/acoustic_model//list/full.lst.

 
I managed to build voices using HTS2.2 but only from small subsets of larger corpora( the corpora used consisted of high quality recordings downsampled to 16 kHz and 16 bit per sample) but when I increased the dimension of the subsets the above presented error occurred. As a note I am not using STRAIGHT for the vocoding, but I consider that to be irrelevant since I got the same error even when using STRAIGHT.

I'm stuck at this point so any help/suggestions will be welcomed. And if you need more info regarding the setup please let me know.

Kindest regards,
Andrei Barbos