1. DSP homework 1 HMM Training and Testing
Advisor: Lin-Shan Lee
Date:

2. Digit Recognizer Construct a Digit Recognizer Free tools of HMM: HTK
Training data, testing data, and some script.

3. Digit Recognizer - Flowchart

4. Thanks for HTK!
We will focus on these HTK tools

5. HCopy - Feature extraction
HCopy -C lib/hcopy.cfg –S scripts/training_hcopy.scp
Convert wave to 39 dimension MFCC
-C lib/hcopy.cfg
Input and output format
e.g. wav -> MFCC_Z_E_D_A
Parameters of feature extraction
Chapter 7 - Speech Signals and Front-end Processing
–S scripts/training_hcopy.scp
A mapping from Input file name to output file name
speechdata/training/N wav MFCC/training/N mfc …

6. Initial MMF prototype MMF: HTK chapter 7
~o <VECSIZE> 39 <MFCC_Z_E_D_A>
~h "proto"
<BeginHMM>
<NumStates> 5
<State> 2
<Mean> 39 …
<Variance> 39 …
<State> 3
<Mean> 13
<Variance> 13 …
<TransP> 5
<EndHMM>

7. HCompV - Initialize Compute global mean and variance of features
HCompV -C lib/config.fig -o hmmdef -M hmm -S scripts/training.scp lib/proto
Compute global mean and variance of features
-C lib/config.fig
Set format of input feature. E.g. MFCC_Z_E_D_A
-o hmmdef -M hmm
Set output name: hmm/hmmdef
-S scripts/training.scp
A list of training data.
lib/proto
A description of a HMM model. HTK MMF format

8. HCompV - Initialize

9. Initial HMM bin/macro bin/models_1mixsil
hmm/models
bin/macro
Construct each HMM
bin/models_1mixsil
Add silence HMM
These are written in C. But you can do these by a text editor.
hmm/hmmdef

10. HERest - Adjust HMM model
HERest -C lib/config.cfg –S scripts/training.scp –I labels/Clean08TR.mlf -H hmm/macros -H hmm/models –M hmm lib/models.lst
Adjust parameters  to maximize P(O|)
one iteration of EM algorithm
Run this command three times => three iteration
–I labels/Clean08TR.mlf
Set label file to “labels/Clean08TR.mlf”
lib/models.lst
A list of word models. ( liN (零), #i (一), #er (二),… jiou (九), sil )

11. HERest - Adjust HMM model
Chapter 4 - basic problem 3 for HMM
Give O and an initial model =(A,B,), adjust  to maximize P(O|)

12. HHEd – Modify HMMs lib/sil1.hed lib/models_sp.lst
bin/spmodel_gen hmm/models hmm/models
Add SP (short pause) HMM definition to MMF file “hmm/models”
HHEd -H hmm/macros -H hmm/models –M hmm
lib/sil1.hed lib/models_sp.lst
lib/sil1.hed
A list of command to modify HMM definitions
lib/models_sp.lst
A new list of model.
( liN (零), #i (一), #er (二),… jiou (九), sil, sp )
See HTK book (p. 33)

13. HERest - Adjust HMM model again
HERest -C lib/config.cfg –S scripts/training.scp
–I labels/Clean08TR_sp.mlf -H hmm/macros
-H hmm/models –M hmm lib/models_sp.lst

14. HHEd - Increase numbers of mixture
HHEd -H hmm/macros -H hmm/models -M hmm lib/mix2_10.hed lib/models_sp.lst
Increase numbers of mixture

15. HERest - Adjust HMM model again
HERest -C lib/config.cfg –S scripts/training.scp
–I labels/Clean08TR_sp.mlf -H hmm/macros
-H hmm/models –M hmm lib/models_sp.lst

16. Flowchart

17. HParse – construct word net
HParse lib/grammar_sp lib/wdnet_sp
lib/grammar_sp
Regular expression
Lib/wdnet_sp
Output Word net

18. HVite – Viterbi search
HVite -H hmm/macros -H hmm/models -S scripts/testing.scp -C lib/config.cfg -w lib/wdnet_sp -l '*' -i result/result.mlf -p 0.0 -s 0.0 lib/dict lib/models_sp.lst
Chapter Search Algorithms for Speech Recognition
-w lib/wdnet_sp
Input word net
-i result/result.mlf
Output MLF file
lib/dict
Dictionary: a mapping from word to phone sequences
E.g. ling -> liN, er -> #er, … . 一 -> sic_i i, 七-> chi_i i

19. HResult – compare with answers
HResults -e "???" sil -e "???" sp -I labels/answer.mlf lib/models_sp.lst result/result.mlf
Longest Common Subsequence (LCS)
====================== HTK Results Analysis ====================
Date: Mon Apr 9 19:34:
Ref : labels/answer.mlf
Rec : result/result.mlf
Overall Results
SENT: %Correct=91.88 [H=441, S=39, N=480]
WORD: %Corr=98.56, Acc=97.41 [H=1713, D=17, S=8, I=20, N=1738]
============================================================

20. Training flowchart

21. Requirement 1 – run baseline
Download homework package
Set PATH for HTK tools
execute (bash shell script):
01_run_HCopy.sh
02_run_HCompV.sh
03_training.sh
04_testing.sh
You can find acc in result/accuracy

22. Requirement 2 – improve recognition accuracy

23. Submit your homework Request files: Deadline: 2007.4.25 0:00
result/accuracy – the best acc you have done
hmm/models – your HMM models
Any script you wrote or modified
A document describe your training process and accuracy
Compress all files above into a zip file named “hw1_b9xxxxxx.zip”
to TA with subject: [DSP] hw1_b9xxxxx <your name>
and your zip file
Deadline: :00

24. Suggestion and other information
Read HTK book chapter 3 first. (except 3.3, 3.5)
Many hints are in the shell scripts.
Perl or a good editor is your good friend.
If you have problems about bash shell script, see
If you want to run HTK under windows
Cygwin + HTK windows binary.
If you have any question, mail to TA
呂東烜