The goal of Exercise 2 is to do more exercises in Tcl/Tk and to introduce you to the Janus objects Database, Dictionary, Tags, Phones and PhonesSet.

Pronunciation Dictionary

Typically, a pronunciation dictionary consists of two fields per entry. The first field gives a word, the second field describes the words pronunciation in terms of phonemes or other units e.g. demi-syllables. Because we are developing a Mandarin speech recognition system, we will use demi-syllables. How to do this is part of the homework.

To create a Dictionary object we first have to create a Tags and Phones object.

Tags are used to add more information to phonemes. We mainly use them to tag phones at the boundary of words. You could use them also for syllable boundaries or tonal information.

We will use the tag WB for word boundaries and T1 T2 T3 T4 T5 for the five different tones used in Mandarin.

To create a Tag object start janus in your steps directory and type:

> Tags tags       ;# create an instance tags of class Tags
> tags puts       ;# output content of the tags instance

> tags add WB     ;# add a tag for word boundaries
> tags add SMOKER

> tags puts       ;# what tags do we have?
SMOKER WB

;# add all the tonal tags with a loop
> foreach tag {T1 T2 T3 T4 T5} {
>   tags add $tag
> }

> tags puts 
SMOKER T1 T2 T3 T4 T5 WB

# show sub-objects that can be accessed
> tags.
item(0..6) list

# we store 7 items starting with an index 0 up to 6 (0..6)
# how can we access these items?

# access item 4
> tags.item(4) puts
T3

> tags delete SMOKER ;# we can also remove tags

> tags.item(4) puts
T4
# But the place/index of items may move if we delete something!

> tags write ./tags.desc ;# creates an ASCII file with the tags
# take a look at the created file

Check the created tags.desc file.

To create a pronunciation dictionary object we have to define the basic acoustic units used to describe the "words". We will now create a phones object. This is only an example! You have to come up with the full list of units for yourself.

The Phones object has a similar functionality as the Tags object.

> Phones phones
> phones add SIL            ;# SIL stand for Silence
> phones add @              ;# This is the padding phoneme. It has a special meaning that we will not discuss now.
> phones add "zh ong b iao" ;# this is a short version to add 4 units at once
> phones puts
@ SIL b iao ong zh

# write the description into an ASCII file
> phones write ./example-phones.desc

# Look at the created file

Now let us create a pronunciation dictionary!

# get some help what other objects are needed
> Dictionary dict -help
ERROR   itf.c(0406) <ITF,FCO>  Failed to create 'dict' object.
Options of 'dict' are:
 <name>    name of the dictionary (string:"dict")
 <Phones>  phones (Phones:)
 <Tags>    tags (Tags:)

# The "ERROR   itf.c(0406)"  is because we did not provide enough information to create the object.
# Additional to the instance name we have to provide first an object of type Phones and then an object of type Tags

# with objects we have created above
> Dictionary dict phones tags

> dict add -help ;# we have to provide two parameters the lexical form and a list list of {phoneme tag*} elements
dict add -help
Options of 'add' are:
 <name>           name (spelling) of the word (string:"NULL")
 <pronunciation>  pronunciation of the word

> dict add zhong3biao2 { {zh WB} {ong T3} b {iao WB T2} }

> dict add $ { {SIL WB} } ;# we will use this word as optional silence, this will be explained later

> dict write ./example.dict ;# write an ASCII version of the dictionary

Take a look at the example.dict file.

Task 3 Write a janus-script that creates a training dictionary (train.dict). (For details see homework).

Now we take a closer look at the PhoneSet object.

Because we will model initials and finals with a different number of HMM states. Therefore, we will group our phonemes.

> PhonesSet phonesSet

# The list of PHONES, INITIALS etc. is incomplete!
> phonesSet add PHONES "SIL @ zh ong b iao"
> phonesSet add INITIALS "zh b"
> phonesSet add FINALS "ong iao"
> phonesSet add SILENCE "SIL"

# We can get a list of elements stored in the set
>  phonesSet
FINALS INITIALS PHONES SILENCE

# We can access the content of e.g. the finals by typing
> phonesSet:FINALS
iao ong

# The type method tells us the type of an janus object
> phonesSet:FINALS type
Phones

# The type of the phonesSet instance is of course
> phonesSet type
PhonesSet

# Write an ASCII description into a file
> phonesSet write example-phonesSet.desc

# Look at the created file

> Dictionary dict2 phonesSet:PHONES tags

> DBase                                ;# shows you the methods defined for the objects of type DBase (method puts)
> DBase db                             ;# create an DBase object db
> db open db.dat db.idx -mode rwc      ;# opens database with data and index file
> db add spk030_utt001 "{ADC  CH030/CH030_1.adc.shn} {TEXT guo2wu4yuan4 zhao4kai1 di4}"           ;#  add an entry with two fields with entry-key = "spk030_utt001"
> db add spk030_utt002 "{ADC  CH030/CH030_2.adc.shn} {TEXT li3peng2 zong3li3 jin1tian1 zhu3chi2}" ;#  add another entry to the database
> db close                             ;# close the database (incl. Files)

> DBase mybase
> mybase open db.dat db.idx -mode r
> makeArray uttInfo [mybase get spk030_utt002]
> puts $uttInfo(TEXT)
li3peng2 zong3li3 jin1tian1 zhu3chi2

[DBase db] open db.dat db.idx -mode rwc
foreach f {[glob  ./data/CH/rmn/*.rmn]
  set fp [open $f r]
  while { [gets $fp line] != -1 } {
    # analyze the text etc.
    set utt ...
    db add ... 
  }
  close $fp
}
db close
exit

         #Deletions + #Substitutions + #Insertions
WER =   -------------------------------------------  * 100
               # words to be recognized

REF: This great machine can recognize       speech
HYP: This       machine can wreck     nice  beach
          DEL               SUB       INS   SUB                       

       1  + 2 + 1
WER = ------------ * 100 = 66  
            6