# -*- coding: utf-8 -*-
"""
Example LandKit Script
Created on Thu Jan 21 15:48:07 2016
@author: jrkerlin
"""
import numpy as np
from matplotlib.pyplot import *
import landkit
reload(landkit)
lktable = landkit.LoadDict("test.csv")
sc = landkit.SentCompare(lktable['SentWords'], lktable['OrigResponse'])
sc.SpellCorrect()
sc.ScoreWords()
sc.GeneratePhonemes()
sc.ScorePhonemes()
sc.GeneratePhonemeTable()
sc.SentenceAnalysis()
#for snum,bools in enumerate(sc.hits_phonemes):
# print sc.target_phonemes[snum]
# print sc.source_phonemes[snum]
# print bools
#
#sentenceACC = []
#for snum,bools in enumerate(sc.hits_phonemes):
# print sc.target[snum]
# print sc.source[snum]
# booze = [int(boo) for boo in bools]
for subjectDir in folders:
blockFolder = normjoin(dataDir, subjectDir)
csvNames = [
normjoin(blockFolder, f) for f in os.listdir(blockFolder)
if fnmatch.fnmatch(f, '*.csv')
]
for section in csvNames:
lks = pd.read_csv(section)
lktable = lktable.append(lks)
print "Now Loading Subject..." + " ".join([subjectDir])
lktable = lktable.reset_index()
# Set all non-responses to a single black
lktable['SourceSentence'] = lktable['SourceSentence'].fillna(" ")
sc = landkit.SentCompare(list(lktable['TargetSentence']),
list(lktable['SourceSentence']), True,
normjoin('C:\Experiments\JK302'))
#filehandler = open(dataDir + '\\sc.pickle', 'w')
#pickle.dump(sc, filehandler)
sc.phonTable.to_csv(dataDir + '\\phonTable.csv')
# Make pandas table
sentT = pd.DataFrame(lktable)
#Join the ngram and IPHoD info and "correctness" at the word level
wordT = pd.concat([sc.tngram, sc.tphod], axis=1, join_axes=[sc.tngram.index])
#
#Table with phoneme Level info
phonT = sc.phonTable
bigPhonT = pd.concat([
#volSent = np.append(volSent,pd.DataFrame(phon,onset,offset,vol))
print(folder + ' ' + fname + ' Volume Tagging...')
#volGrouped.append(volSent)
tt['FileID'] = fCount
tt['SpeechRMS'] = volume
audioTable = pd.concat([audioTable, tt])
fCount += 1
audioTable.to_csv(os.path.normpath(r'C:\TCDTIMIT\Tables\audioTable_r1.csv'))
#Match HTK forced alignments to landkit phonemes
taggedCorpus = pd.DataFrame.from_csv(normjoin(outPath, 'posTags_r1.csv'))
justSent = taggedCorpus.groupby(['Subject', 'File']).first().reset_index()
sc = landkit.SentCompare(list(justSent['Sentence']),
list(justSent['Sentence']), False,
normjoin('C:\Experiments\JK302'))
sc.GeneratePhonemes()
audioTable = pd.DataFrame.from_csv(
os.path.normpath(r'C:\TCDTIMIT\Tables\audioTable_r1.csv'))
def source2target(sourceList, targetList, sourceVal, fillOpt='Neg1'):
import difflib
d = difflib.Differ()
list(d.compare(sourceList, targetList))
instruct = [x[0] for x in list(d.compare(sourceList, targetList))]
c = 0
print "Now Loading ..." + " ".join([folder, folder2])
lktable = lktable.reset_index()
speechToText = []
for f in lktable['VerbalResponse']:
try:
transcript = landkit.ATTSR(f)
except:
pass
transcript = " "
speechToText.append(transcript)
print speechToText[-1]
lktable['speechToText'] = speechToText
# Set all non-responses to a single black
lktable['SourceSentence'] = lktable['SourceSentence'].fillna(" ")
sc = landkit.SentCompare(list(lktable['TargetSentence']),
list(lktable['SourceSentence']), True)
filehandler = open(dataDir + '\\sc.pickle', 'w')
pickle.dump(sc, filehandler)
sc.phonTable.to_csv(dataDir + '\\phonTable.csv')
# Make pandas table
sentT = pd.DataFrame(lktable)
#Join the ngram and IPHoD info and "correctness" at the word level
wordT = pd.concat([sc.tngram, sc.tphod], axis=1, join_axes=[sc.tngram.index])
#wordT['HitIndex']= [y for x in sc.hits for y in x]
#wordT['WordIndex']= [y for x in sc.hits for y,z in enumerate(x)]
#
import numpy as np
from matplotlib import *
import sys
#Add path to landkit
sys.path.append(r'C:\Users\jrkerlin\Documents\GitHub\landk\Analysis')
import landkit
reload(landkit)
##Just the spell correction and word -level scoring
#sc = landkit.SentCompare(['hello worlddsfg go','soul train with him'],['hellot mold','sol tarin'],False)
#sc.SpellCorrect()
#sc.ScoreWords()
lktable = pd.read_csv(r'C:\TCDTIMIT\dataOut\sb\1456357335\sb1456357335.csv')
sc = landkit.SentCompare(lktable['TargetSentence'], lktable['SourceSentence'],
False)
## Make pandas table
#sentT = pd.DataFrame(lktable)
#
##Join the ngram and IPHoD info and "correctness" at the word level
#wordT = pd.concat([sc.tngram, sc.tphod], axis=1, join_axes=[sc.tngram.index])
#wordT['HitIndex']= [y for x in sc.hits for y in x]
#wordT['WordIndex']= [y for x in sc.hits for y,z in enumerate(x)]
#
##Table with phoneme Level info
#phonT = sc.phonTable
#bigPhonT = pd.concat([landkit.IndexFill(sentT,phonT['SentenceCount'],phonT['PhonemeCount']),landkit.IndexFill(wordT,phonT['WordCount'],phonT['PhonemeCount']),phonT], axis=1, join_axes=[phonT.index])
#bigPhonT = pd.concat([landkit.IndexFill(wordT,phonT['WordCount'],phonT['PhonemeCount']),phonT], axis=1, join_axes=[phonT.index])
#bigPhonT.to_csv('C:\TCDTIMIT\DataMine\bigP.csv')
#wordT.to_csv('C:\TCDTIMIT\DataMine\wordT.csv')
print(textIn)
else:
print('user cancelled')
#Record final response
if textIn:
sourceSentence = textIn[0]
else:
sourceSentence = " "
print sourceSentence
print "target " + targetSentence
table['SourceSentence'][trial] = sourceSentence
table['TargetSentence'][trial] = targetSentence
#Just the spell correction and word -level scoring
sc = landkit.SentCompare([targetSentence], [sourceSentence], False)
#sc.SpellCorrect() enchant crashes this script at testing computer for unknown reason!!!
sc.ScoreWords()
wscore = sc.wscore
print wscore
table['SpellCorrSource'][trial] = sc.source[0]
table['SentenceWordScore'][trial] = wscore[0]
#Adapt dbSNR on every trial
if wscore > 50:
dBSNR += -3
elif wscore == 50:
dBSNR += 0
for subjectDir in folders:
blockFolder = normjoin(dataDir, subjectDir)
csvNames = [
normjoin(blockFolder, f) for f in os.listdir(blockFolder)
if fnmatch.fnmatch(f, '*.csv')
]
for section in csvNames:
lks = pd.read_csv(section)
lktable = lktable.append(lks)
print "Now Loading Subject..." + " ".join([subjectDir])
lktable = lktable.reset_index()
# Set all non-responses to a single black
lktable['SourceSentence'] = lktable['SourceSentence'].fillna(" ")
sc = landkit.SentCompare(list(lktable['FullSentence']),
list(lktable['SourceSentence']), True,
normjoin('C:\Experiments\JK310'))
#filehandler = open(dataDir + '\\sc.pickle', 'w')
#pickle.dump(sc, filehandler)
sc.phonTable.to_csv(dataDir + '\\phonTable.csv')
# Make pandas table
sentT = pd.DataFrame(lktable)
#Join the ngram and IPHoD info and "correctness" at the word level
wordT = pd.concat([sc.tngram, sc.tphod], axis=1, join_axes=[sc.tngram.index])
#
#Table with phoneme Level info
phonT = sc.phonTable
bigPhonT = pd.concat([
