[hts-users:04249] Re: Parallel mode HERest in HTS

I found the reason is that there should be two phone lists (cmp model and duration model) in the HERest command.

so the problem can be fixed using the command as below:

HERest -p 0 -A -B -C ./configs/qst001/ver1/trn.cnf -D -T 1 -I ./data/labels/full.mlf -m 1 -u tmvwdmv -w 5000 -t 1500 100 5000 -H ./models/qst001/ver1/cmp/clustered.mmf -N ./models/qst001/ver1/dur/clustered.mmf -M ./models/qst001/ver1/cmp -R ./models/qst001/ver1/dur ./data/lists/full.list ./data/lists/full.list ./models/qst001/ver1/cmp/HER1.hmm.acc ./models/qst001/ver1/cmp/HER1.dur.acc ./models/qst001/ver1/cmp/HER2.hmm.acc ./models/qst001/ver1/cmp/HER2.dur.acc

and as Truong Do said, the "-I mlf" option is useless.

But I think "-C config" should keep in the command.

So the command finally looks like:

HERest -p 0 -A -B -C ./configs/qst001/ver1/trn.cnf -D -T 1 -m 1 -u tmvwdmv -w 5000 -t 1500 100 5000 -H ./models/qst001/ver1/cmp/clustered.mmf -N ./models/qst001/ver1/dur/clustered.mmf -M ./models/qst001/ver1/cmp -R ./models/qst001/ver1/dur ./data/lists/full.list ./data/lists/full.list ./models/qst001/ver1/cmp/HER1.hmm.acc ./models/qst001/ver1/cmp/HER1.dur.acc ./models/qst001/ver1/cmp/HER2.hmm.acc ./models/qst001/ver1/cmp/HER2.dur.acc

BTW: the order of hmm.acc and dur.acc is important.

the correct order is: .... HER1.hmm.acc HER1.dur.acc HER2.hmm.acc HER2.dur.acc HER3.hmm.acc HER3.dur.acc.....

On Sat, Mar 28, 2015 at 9:01 AM, Truong Do <truongdq54@gmail.com> wrote:

Hi Qingsong Liu,

I'm new to HTS-user mailing list, so I don't know how to post the answer there.

For your problem, when running the final HERest with -p 0, you don't have to
put "-C" or "-I" options.

Please takes a look at this page:
https://truongdq.com/index/hts-parallel-training

Hope this help!
Best,
Truong

Hi,

I use parallel mode of HERest in HTS (option -p, which can dump the accumulators to disk and then reload them later).

but the error is as below when merge the accumulators with -p 0 :

Merging results:

HERest -p 0 -A -B -C ./configs/qst001/ver1/trn.cnf -D -T 1 -I ./data/labels/full.mlf -m 1 -u tmvwdmv -w 5000 -t 1500 100 5000 -H ./models/qst001/ver1/cmp/clustered.mmf -N ./models/qst001/ver1/dur/clustered.mmf -M ./models/qst001/ver1/cmp -R ./models/qst001/ver1/dur ./data/lists/full.list ./models/qst001/ver1/cmp/HER1.hmm.acc ./models/qst001/ver1/cmp/HER1.dur.acc ./models/qst001/ver1/cmp/HER2.hmm.acc ./models/qst001/ver1/cmp/HER2.dur.acc

HTK Configuration Parameters[8]

Module/Tool Parameter Value

# MINDUR 5

# MAXSTDDEVCOEF 10

# APPLYDURVARFLOOR TRUE

# DURVARFLOORPERCENTILE 1.000000

# VFLOORSCALESTR Vector 4 0.01 0.01 0.01 0.01

# NATURALWRITEORDER TRUE

# NATURALREADORDER TRUE

# APPLYVFLOOR TRUE

HMM:

HERest ML Updating: Transitions Means Variances MixWeights

Parallel-Mode[0] System is SHARED

276774 Logical/276774 Physical Models Loaded, VecSize=108

1 MMF input files

ERROR [+7060] InitHMMSet: Expected newline after 2'th HMM

ERROR [+2321] Initialise: MakeHMMSet failed

FATAL ERROR - Terminating program HERest

parallel mode

operation of HERest (option -p, which can dump the accumulators

to disk and then reload them later)

could anyone help me to fix this problem?

Thank you.

--

---------------------------------------------
Qingsong Liu
liuqs.ustc@gmail.com
Univ. of Sci.& Tech. of China
----------------------------------------------

---------------------------------------------
Qingsong Liu
liuqs.ustc@gmail.com
Univ. of Sci.& Tech. of China
----------------------------------------------