georgid / lyrics2audioaligner Goto Github PK

In essence they are different in

• the grammar table in Aligner,
• the grapheme2phoneme lookup table in Phonetizer
  --- these two tables cn abe put into a txt file?
• doitOneChunk, where the resynthesized file is directly loaded. .wav
  this file is full of garbage. clean it!

put it in method evaluation.TextGrid_Parsing.TextGrid2WordList()

see post
http://www.gossamer-threads.com/lists/python/python/595835

remove dividebyLineAllLyrics method.
leave makamScore.serializeLyricsToFile() for checking in result in praat for each line.

Best is to add in prepareOutputForPRaat logic for 1) phonemeAligned and 2) for wordAligned.

NOTE: in praat script add one layer for each with the etensioans given above.

NAMING CONVENTION:
rename phonemeAligned to phonemeAlignedMLF
Then rename the shitty phonemeAlignednoMLF to phonemeAligned

change

• Phonetizer.METULyrics2phoneticDict
• aligner._toWordNetwork
• Aligner.Aligner._createWordMLFandDict

APROPO: in the last one:
METULyrics = Phonetizer.turkishScriptLyrics2METUScriptLyricsFile(baseNameAudioFile + LYRICS_TXT_EXT, METUBETfileName)
is this needed for HTK? I think no, because it works in the varnam branch.

but first get rid of METU-layer methods in Phonetizer
check how methods from phonetizer look like in Varnam-Branch

steps needed:

• find for which compositions scores are available in symbTr2.0 https://github.com/MTG/SymbTr/blob/master/txt
• parse pauses and calc durations based on pauses.
• make new .sectionAnnotations

in class Adapt.adapt the MAP logic is done in python.

So left is to put the logic for MLLR in script.

Take the business logic away form main and put it in doit method.

WordLevelAval.evalPhraseLevelError

start with line 47:
It is duplicate functionality.

This will spare the step of transforming turkish to intermediate and then to METU BET.
So this way all annotations .TextGrid should be rewritten to turkish script.

As well decide if we leave the system to work with .txt files or use only .TextGrid. This might not be good idea for use cases when no TextGrid is available. Where is the load of textGrid instead of .txtTur handled in varnam_noSynth?

/Users/joro/Documents/Phd/UPF/voxforge/myScripts/AlignmentStep/doForceAligment.sh: line 113: 98769 Abort trap $HTK_34_PATH/HVite -l "'*'" -o SW -A -D -T 1 -b sil -C $PARENT_OF_INTERIM_AND_INPUT_FILES/input_files/config -a -H $HMM -i $PHONE_LEVEL_ALIGNMENT -m -I $WORD_LEVEL_MLF -y lab /tmp/lexicon2 $HMMLIST ${1}.mfc

for branch words : MakamScore.MakamScore.getIndicesPhonemes and MakamScore.MakamScore.getIndicesPhonemes_durations could be combined.

Is it possible to switch to one table and redo all TexGrid annotations with turkish script