Practice Problem 2:

Use any or all of bash, sed, awk, perl, python, or ruby to create
auxliary htk files for one or more of the available corpora mentioned
in Lecture 1, or for a corpus of your choice.

For example, if you are working with the AVICAR Corpus, create: 

1.  The master label file (mlf) without time alignments for files containing 
TIMIT sentences and digits.  You will need to create these 
files using the file AVICAR.txt.  This file contains two kinds of entries

D##_###_P0_C1   (9 2 6) 4 8 7 - 4 0 8 4

and

D##_##_S4_C2   That dog chases cats mercilessly.

The first letter in the filename specifies which block of utterances 
the audio files are from.  There are 10 blocks.  Each block is labeled A-J.  
The second letter in the file name (M|F) indicates the gender of the speaker.  
The third character in the file name indicates the speaker number.  
There are a total of 10 speakers in each block, 5 male and 5 female.  
The middle block of characters indicate the noise conditions in 
the file.  There are 5 noise conditions.  They are car idling (IDL), car going 
35 MPH with the windows up (35U), car going 35 MPH with the windows down (35D), 
car going 55 MPH with the windows up (55U), and car going 55 MPH with the windows 
down (55D).   

In addition, each utterance can be from one of 8 microphone channels, so in order to 
specify which channel, each filename will need an _M(1-8).  An example of an mlf file
created from the two above utterances is:

#!MLF!#
"*/DM1_IDL_P0_C1_M1.lab''
nine
two
six
four
eight
seven
four
zero
eight
four
.
"*/DM1_IDL_P0_C1_M2.lab''
nine
two
six
four
eight
seven
four
zero
eight
four
.
"*/DF3_55D_S4_C2_M5.lab"
That
dog 
chases 
cats 
mercilessly
.
"*/DM4_35U_S4_C2_M8.lab"
That
dog 
chases 
cats 
mercilessly
.

2.  The script file that HTK will use to locate the acoustic feature
files.  This file should contain entries of the form

/homes/sborys/AVICAR/htk_mfcc/DM1_IDL_P0_C1_M1.mfc
/homes/sborys/AVICAR/htk_mfcc/DM1_IDL_P0_C1_M2.mfc
/homes/sborys/AVICAR/htk_mfcc/DF3_55D_S4_C2_M5.mfc
/homes/sborys/AVICAR/htk_mfcc/DM4_35U_S4_C2_M8.mfc

3.  The script file that will tell HCopy what names to use when
transforming audio files into acoustic feature files.  This file
should have entries in the following format:

/homes/sborys/AVICAR/audio/DM1_IDL_P0_C1_M1.wav /homes/sborys/AVICAR/htk_mfcc/DM1_IDL_P0_C1_M1.mfc
/homes/sborys/AVICAR/audio/DM1_IDL_P0_C1_M2.wav /homes/sborys/AVICAR/htk_mfcc/DM1_IDL_P0_C1_M2.mfc
/homes/sborys/AVICAR/audio/DF3_55D_S4_C2_M5.wav /homes/sborys/AVICAR/htk_mfcc/DF3_55D_S4_C2_M5.mfc
/homes/svorys/AVICAR/audio/DM4_35U_S4_C2_M8_wav /homes/sborys/AVICAR/htk_mfcc/DM4_35U_S4_C2_M8.mfc

4.  Lists of words in the corpus.  Each word need only be listed once.
For example

cat
chased
dog
four
one
the
three
two 

For the UASpeech corpus, create the four kinds of files listed above.
Audio filenames in the UASpeech corpus are in the format

M16_B2_CW8_M5.wav

The first three characters in the filename indicate which speaker
produced the utterance.  M16 means that this recording was produced by
male speaker number 16.  The next set of two characters indicates the
block number.  There are three blocks, B1, B2, and B3.  The next part
of the filename specifies what kind of utterance is in the file.  CW8
indicates that this file contains speaker M16's production of Common
Word number 8.  The last lump of characters in the filename indicates
which microphone channel was used.  There were 7 microphone channels.

To create mlf files for this corpus, you will need the auxilary file
UASpeech.txt.  This file contains entries like.

oil	CW86
equilibrium	B1_UW35

The first column is the word utterance contained in the audiofile.
The second column is can be used to reconstruct the UASpeech audio
filenames.  Entries in column 2 that do not contain a B? indicator are
included in all three blocks so you will need to create a label file
for each block.  Create label files for speakers M01, M04, M05, M06,
M07, M08, M09, M10, M12, M13, M14, M15, M16, F02, F03, and F04 for all
seven microphone channels for all three blocks of data.

Both script files will have the same file format as those shown above.
The list of words will also have the same format shown above.

Create the same four files for the Buckeye and AMI corpora.  For these
corpora, you will need to get the documentation available online.

Feel free to email or visit Sarah (sborys@uiuc.edu, 2013 Beckman) or
HJ (jhasegaw@uiuc.edu, 2011 Beckman) with questions, comments,
concerns, complaints, complements regarding this ``assignment''.