Step 6 - Creating Flat Start Monophones

Define Prototype Model

The first step in Hidden Markov Model ("HMM") training is defining a prototype model called "proto". The focus here is to create a model structure, the parameters are not important. Create a file called 'proto' in your 'voxforge/manual' directory and add the following:

~o <VecSize> 25 <MFCC_0_D_N_Z>
~h "proto"
<BeginHMM>
<NumStates> 5
<State> 2
    <Mean> 25
      0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
    <Variance> 25
      1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
<State> 3
    <Mean> 25
      0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
    <Variance> 25
      1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
<State> 4
    <Mean> 25
      0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
    <Variance> 25
      1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
<TransP> 5
0.0 1.0 0.0 0.0 0.0
0.0 0.6 0.4 0.0 0.0
0.0 0.0 0.6 0.4 0.0
0.0 0.0 0.0 0.7 0.3
0.0 0.0 0.0 0.0 0.0
<EndHMM>

For details of what all this means, see the HTK book.

You also need a configuration file. Create a file called 'config' in your 'voxforge/manual' directory and add the following data:

TARGETKIND = MFCC_0_D_N_Z
TARGETRATE = 100000.0
SAVECOMPRESSED = T
SAVEWITHCRC = T
WINDOWSIZE = 250000.0
USEHAMMING = T
PREEMCOEF = 0.97
NUMCHANS = 26
CEPLIFTER = 22
NUMCEPS = 12

Note: the target kind in you proto file (the "MFCC_0_D_N_Z" on the first line), needs to match the TARGETKIND in your config file.

You also need to tell HTK where all your feature vector files are located (those are the mfcc files you created in the last step). You do this with with a script file. Therefore, create a file called "train.scp".

The next step is to create a new folder call hmm0. Then create a new version of proto in the hmm0 folder - using the HTK HCompV tool as follows:

$HCompV -A -D -T 1 -C config -f 0.01 -m -S train.scp -M hmm0 proto

This creates two files in the hmm0 folder:

Flat Start Monophones

create hmmdefs

Create a new file called hmmdefs in your 'voxforge/manual/hmm0' folder:

Copy the monophones0 file to your hmm0 folder;
rename the monophones0 file to hmmdefs;

For each phone in hmmdefs:

put the phone in double quotes;

add '~h ' before the phone (note the space after the '~h'); and

copy from line 5 onwards (i.e. starting from "<BEGINHMM>" to "<ENDHMM>") of the hmm0/proto file and paste it after each phone.

Leave one blank line at the end of your file.

This create the hmmdefs hmmdefs file, which contains "flat start" monophones.

Create macros File

The final step in this section is to create the macros file.

A new file called macros should be created and stored in your 'voxforge/manual/hmm0' folder:

create a new file called macros in hmm0;
copy vFloors to macros
copy the first 3 lines of proto (from ~o to <DIAGC>) and add them to the top of the macros file

It should look something like this when you have finished:

~o
<STREAMINFO> 1 25
<VECSIZE> 25<NULLD><MFCC_D_N_Z_0><DIAGC>
~v varFloor1
<Variance> 25
6.580433e-01 3.732679e-01 3.525512e-01 4.770427e-01 4.332326e-01 4.544643e-01 5.620688e-01 2.553866e-01 4.001570e-01 3.416672e-01 2.128212e-01 2.660224e-01 1.668585e-02 1.700366e-02 1.616409e-02 1.768895e-02 1.718035e-02 2.098122e-02 2.326025e-02 1.677738e-02 2.010740e-02 1.595870e-02 1.417547e-02 1.510511e-02 1.447709e-02

Re-estimate Monophones

Next, create 9 new folders named consecutively in your 'voxforge/manual' folder: hmm1 to hmm9.

The Flat Start Monophones are re-estimated using the HERest tool. The purpose of this is to load all the models in the hmm0 folder (these are contained in the hmmdefs file), and re-estimate them using the MFCC files listed in the train.scp script, and create a new model set in hmm1. Execute the HERest command from your 'voxforge/manual' directory:

$HERest -A -D -T 1 -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H hmm0/macros -H hmm0/hmmdefs -M hmm1 monophones0

The files created by this command are:

This process is repeated 2 more times, creating new model sets in hmm2 and hmm3, respectively:

$HERest -A -D -T 1 -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H hmm1/macros -H hmm1/hmmdefs -M hmm2 monophones0

The files created by this command are:

$HERest -A -D -T 1 -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H hmm2/macros -H hmm2/hmmdefs -M hmm3 monophones0

The files created by this command are:

Comments

Click the 'Add' link to add a comment to this page; click the 'Read More' link to view replies to a posted comment.

Add • Search

More info on Feature Parameters

By kmaclean - 9/23/2008

The Julius 3.2 book section 7. Feature parameter files (-input mfcfile) says:

Feature Parameter Types
It is necessary that the feature parameter format matches the original HMM training data feature format. However if all the necessary parameters for the HMM are held within the given feature parameter file,Julius will then automatically extract the appropriate parameters for recognition.

For example
If the parameter format below is used for training
      MFCC_E_D_N_Z = MFCC(12)+ ?MFCC(12)+ ?Pow(1) (CMN)                   25-dimension
Then for recognition you can also use feature parameter files other then MFCC_E_D_N_Z, such as
      MFCC_E_D_Z        = MFCC(12)+Pow(1)+?MFCC(12)+ ?Pow(1)
                                                           (CMN) 26-dimension
                                                                                                      or
     MFCC_E_D_A_Z = MFCC(12)+Pow(1)+ ?MFCC(12)+ ?Pow(1)
                     + ??MFCC(12) + ??Pow(1) (CMN)             39-dimension
The parameter file needs to contain all of the parameters used for the original training of the HMM model,
extra data contained with in the file will not be used.

where (from the HTK book):

_E has energy

_N absolute energy suppressed

_D has delta coe?cients

_A has acceleration coe?cients

_C is compressed

_Z has zero mean static coef.

_K has CRC checksum

_O has 0’th cepstral coef

Error +6510

By dave - 11/13/2008 - 2 Replies

Hello,

I follow the Tutorial step by step

but by running the command:

HERest -A -D -T 1 -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H hmm0/macros -H hmm0/hmmdefs -M hmm1 monophones0I

I get the Error:

HTK Configuration Parameters[10]
Module/Tool     Parameter                  Value
#                 NUMCEPS                       12
#                 CEPLIFTER                     22
#                 NUMCHANS                      26
#                 PREEMCOEF               0.970000
#                 USEHAMMING                  TRUE
#                 WINDOWSIZE         250000.000000
#                 SAVEWITHCRC                 TRUE
#                 SAVECOMPRESSED              TRUE
#                 TARGETRATE         100000.000000
#                 TARGETKIND          MFCC_0_D_N_Z

HERest ML Updating: Transitions Means Variances

System is PLAIN
29 Logical/29 Physical Models Loaded, VecSize=25
2 MMF input files
Pruning-On[250.0 150.0 1000.0]
Processing Data: Sample1.mfc; Label Sample1.lab
ERROR [+6510] LOpen: Unable to open label file ../train_mfc/Sample1.lab
FATAL ERROR - Terminating program HERest

I have build the phone0.mlf and the word.mlf file.

What i also don't understand is why he search in the folder /train_mfc/

there are my mfc files.

I hope someone can help me.

Thanks

Dave

untitled

By Alan - 10/5/2008 - 1 Replies

I had problem with

ERROR [+6510] LOpen: Unable to open label file

I followd through the tutorial and also going through the htk book, but no where during the tutorial has it created any .lab files :S (wasnt it suppose to create mfc at HLEd step?) i tried reading the link u gave to another person with the same question, but i dnt get it, i did the HLEd step corret with .mfc files made and the path correct, but now its just telling me i cannot open label files which was never mentioned b4 :S

so lost...

problem with the HERest

By imene - 8/8/2008 - 1 Replies

hi

I have a problem with the HERest command.

I don't have the description phonetics of each file of my base of data.

am I can execute the comande "HERest" without the files '.lab' because when I try it I had the problem

Processing data: sa1.mfcc; label sa1.lab
error [+6510] LOpen: Unable to open label file dated/train_lab/sa1.lab

help me please

Step 6 - Unknown label error

By crystal - 1/17/2008 - 3 Replies Hi i run the above command

HERest -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H hmm0/macros -H hmm0/hmmdefs -M hmm1 monophones0

But i got an error shows:

Pruning-On[250.0 150.0 1000.0]
ERROR [+7321] CreateInsts: Unknown label YOU

Can anyone help me??thanks


Username	Password

Step 6 - Creating Flat Start Monophones

Define Prototype Model

Flat Start Monophones

create hmmdefs

Create macros File

Re-estimate Monophones

Comments

Create macros File