STEP 7 Fixing The Silence Model

First copy the contents of the hmm3 folder to hmm4
Then using an editor, create new "sp" model in hmm4/hmmdefs as follows:

copy and paste the “sil” model and rename the new one "sp"

remove state 2 and 4 from new “sp” model (i.e. keep 'centre state' of old “sil” model in new "sp" model )

change numstates to 3

change states to 2

change transp to 3

change matrix in transp to 3 by 3 array

change numbers in matrix as follows:

0.0 1.0 0.0
0.0 0.5 0.5
0.0 0.0 0.0
This creates hmmdefs and macros in hmm4 folder.


Next, create the following HHEd command script, called sil.hed.

AT 2 4 0.2 {sil.transP}
AT 4 2 0.2 {sil.transP}
AT 1 3 0.3 {sp.transP}
TI silst {sil.state[3],sp.state[2]}

Next run HHEd as follows, but using the monophones1 file which contains the sp model:
HHEd -H hmm4/macros -H hmm4/hmmdefs -M hmm5 sil.hed monophones1

This creates hmmdefs and macros in hmm5 folder.

Next run HERest 2 more times, this time using the monophones1 file:

HERest -T 1 -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S
train.scp -H hmm5/macros -H hmm5/hmmdefs -M hmm6 monophones1

HERest -T 1 -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S
train.scp -H hmm6/macros -H hmm6/hmmdefs -M hmm7 monophones1

This creates hmmdefs and macros in hmm6 and hmm7 folder as well.

From here onwards our monophones based system is ready for recognition.If we run the command

cmd>HVite -H hmm7/macros -H hmm7/hmmdefs -C config2 -w wdnet dict
monophones1

Creating Monophone HMMs: STEP 6 Creating Flat Start Monophones

In this section, the creation of a well trained monophone HMMs will be described.

In the first step in HMM training we define a prototype model. The parameters of this model are not important, its purpose is to define the model topology.
Create a file called 'proto' in your working directory which is as follows

~o Vecsize 39 MFCC_0_D_A
~h "proto"
BeginHMM
Numstates 5
State 2
Mean 39
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 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 39
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 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 39
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 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 39
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 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 39
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 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 39
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 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

(some tags are not visible. add "<> " tag wherever any word is written)

Now, Create a file called 'config' in your working directory with the following containts

# Coding parameters
TARGETKIND = MFCC_0_D_A
TARGETRATE = 100000.0
SAVECOMPRESSED = T
SAVEWITHCRC = T
WINDOWSIZE = 250000.0
USEHAMMING = T
PREEMCOEF = 0.97
NUMCHANS = 26
CEPLIFTER = 22
NUMCEPS = 12
ENORMALISE = F
(note that this configuration is different from wav_config)
We need to locate all our feature vector files with a script file called train.scp.
mfcc/S0001.mfc
mfcc/S0004.mfc
mfcc/S0005.mfc
mfcc/S0008.mfc
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:

proto

vFloors

Flat Start Monophones

Create hmmdefs

1. Create a new file called hmmdefs in your hmm0 folder:

o Copy the monophones0 file to your hmm0 folder;

o rename the monophones0 file to hmmdefs

2. 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 "" to "") of the hmm0/proto file
and paste it after each phone.

Leave one blank line at the end of your file. This creates the 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 hmm0 folder:

create a new file called macros in hmm0;

copy vFloors to macros

copy the first 3 lines of proto (from ~o to ) and add them to the top of the macros file


Re-estimate Monophones
Next, create 9 new folders named consecutively in your working 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 working directory:

cmd> 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

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

cmd> HERest -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H
hmm1/macros -H hmm1/hmmdefs -M hmm2 monophones0

cmd> HERest -C config -I phones0.mlf -t 250.0 150.0 1000.0 -S train.scp -H
hmm2/macros -H hmm2/hmmdefs -M hmm3 monophones0