def doitOneChunk(argv):
if len(argv) != 8 and len(argv) != 9 :
print ("usage: {} <pathToComposition> <URI_recording_no_ext> <withDuration=True> <withSynthesis> <ALPHA> <ONLY_MIDDLE_STATE> <evalLevel> <usePersistentFiles=True>".format(argv[0]) )
sys.exit();
URIrecordingNoExt = argv[2]
whichSection = getSectionNumberFromName(URIrecordingNoExt)
pathToComposition = argv[1]
withDuration = argv[3]
if withDuration=='True':
withDuration = True
elif withDuration=='False':
withDuration = False
else:
sys.exit("withDuration can be only True or False")
withSynthesis = argv[4]
if withSynthesis=='True':
withSynthesis = True
elif withSynthesis=='False':
withSynthesis = False
else:
sys.exit("withSynthesis can be only True or False")
ALPHA = float(argv[5])
ONLY_MIDDLE_STATE = argv[6]
evalLevel = tierAliases.wordLevel
evalLevel = int(argv[7])
params = Parameters(ALPHA, ONLY_MIDDLE_STATE)
usePersistentFiles = 'True'
if len(argv) == 9:
usePersistentFiles = argv[8]
set_printoptions(threshold='nan')
################## load lyrics and models
htkParser = None
if withDuration:
htkParser = HtkConverter()
htkParser.load(MODEL_URI, HMM_LIST_URI)
alignmentErrors, detectedWordList, grTruthDurationWordList, detectedAlignedfileName = alignDependingOnWithDuration(URIrecordingNoExt, whichSection, pathToComposition, withDuration, withSynthesis, evalLevel, params, usePersistentFiles, htkParser)
mean, stDev, median = getMeanAndStDevError(alignmentErrors)
# writeListOfListToTextFile(detectedWordList, None, '/Users/joro/Downloads/test.txt')
logger.info("mean : {} st dev: {} ".format( mean,stDev))
def loadSmallAudioFragment(lyrics, URIrecordingNoExt, withSynthesis, fromTs=-1, toTs=-1):
'''
test duration-explicit HMM with audio features from real recording and htk-loaded model
asserts it works. no results provided
'''
htkParser = HtkConverter()
htkParser.load(MODEL_URI, HMM_LIST_URI)
lyricsWithModels = LyricsWithModels(lyrics, htkParser, 'False')
observationFeatures = loadMFCCs(URIrecordingNoExt, withSynthesis, fromTs, toTs) # observationFeatures = observationFeatures[0:1000]
lyricsWithModels.duration2numFrameDuration(observationFeatures, URIrecordingNoExt)
# lyricsWithModels.printWordsAndStates()
return lyricsWithModels, observationFeatures
def loadModelsForGivenPhoneme(modelBefore, modelAfter, phonemeName):
# load models
conv_before = HtkConverter()
conv_before.load(modelBefore, HMM_LIST_URI)
conv_after = HtkConverter()
conv_after.load(modelAfter, HMM_LIST_URI)
# get models for given phoneme:
hmmModels = [hmm for hmm in conv_before.hmms if hmm.name == phonemeName]
hmmModel1 = hmmModels[0]
hmmModels = [hmm for hmm in conv_after.hmms if hmm.name == phonemeName]
hmmModel2 = hmmModels[0]
return hmmModel1, hmmModel2
def loadModelsForGivenPhoneme(modelBefore, modelAfter, phonemeName):
# load models
conv_before = HtkConverter()
conv_before.load(modelBefore, HMM_LIST_URI)
conv_after = HtkConverter()
conv_after.load(modelAfter, HMM_LIST_URI)
# get models for given phoneme:
hmmModels = [hmm for hmm in conv_before.hmms if hmm.name == phonemeName]
hmmModel1 = hmmModels[0]
hmmModels = [hmm for hmm in conv_after.hmms if hmm.name == phonemeName]
hmmModel2 = hmmModels[0]
return hmmModel1, hmmModel2
output = output + str(element) + "\t"
output = output.strip()
output = output + '\n'
outputFileHandle.write(output)
outputFileHandle.close()
if __name__ == '__main__':
# findDiffMean(phonemeName)
conv_before = HtkConverter()
conv_before.load(MODEL_URI, HMM_LIST_URI)
# one phoneme
phonemeName = 'A'
hmmModels = [hmm for hmm in conv_before.hmms if hmm.name == phonemeName]
hmmModel1 = hmmModels[0]
# all phonemes
for currHmmModel in conv_before.hmms:
# printModelsMiddleState(currHmmModel, PATH_TO_HMMLIST);
printModelsAllStates(currHmmModel, PATH_TO_HMMLIST);
def doitOneRecording(argv):
'''
for a list of recordings, select those which name contains pattern and evlauate total error
'''
if len(argv) != 9 and len(argv) != 10 :
print ("usage: {} <pathToComposition> <pathToRecordings> <pattern> <withDuration=True/False> <withSynthesis> <ALPHA> <ONLY_MIDDLE_STATE> <evalLevel> <usePersistentFiles=True> ".format(argv[0]) )
sys.exit();
os.chdir(argv[2])
# get annot files with starting pattern
pattern = argv[3] + '*' + AUDIO_EXT
listAudioFilesAll = glob.glob(pattern)
for i in range(len(listAudioFilesAll)) :
listAudioFilesAll[i] = os.path.join(argv[2], listAudioFilesAll[i])
# listAudioFiles = []
# if not isfile( os.path.splitext(listAudioFilesAll[i])[0] + ".notUsed"):
# listAudioFiles.append(listAudioFilesAll[i])
listAudioFiles = listAudioFilesAll
for file in listAudioFiles:
logger.debug(file)
pathToComposition = argv[1]
withDuration = argv[4]
if withDuration=='True':
withDuration = True
elif withDuration=='False':
withDuration = False
else:
sys.exit("withDuration can be only True or False")
withSynthesis = argv[5]
if withSynthesis=='True':
withSynthesis = True
elif withSynthesis=='False':
withSynthesis = False
else:
sys.exit("withSynthesis can be only True or False")
ALPHA = float(argv[6])
ONLY_MIDDLE_STATE = argv[7]
params = Parameters(ALPHA, ONLY_MIDDLE_STATE)
evalLevel = int(argv[8])
usePersistentFiles = 'True'
if len(argv) == 10:
usePersistentFiles = argv[9]
totalErrors = []
htkParser = None
if withDuration:
htkParser = HtkConverter()
htkParser.load(MODEL_URI, HMM_LIST_URI)
for URI_annotation in listAudioFiles :
URIrecordingNoExt = os.path.splitext(URI_annotation)[0]
logger.debug("PROCESSING {}".format(URIrecordingNoExt) )
whichSection = getSectionNumberFromName(URIrecordingNoExt)
currAlignmentErrors, detectedWordList, grTruthDurationWordList, detectedAlignedfileName = alignDependingOnWithDuration(URIrecordingNoExt, whichSection, pathToComposition, withDuration, withSynthesis, evalLevel, params, usePersistentFiles, htkParser)
totalErrors.extend(currAlignmentErrors)
# visualiseInPraat(URIrecordingNoExt, withDuration, detectedWordList, grTruthDurationWordList)
mean = []
stDev = []
if len(totalErrors) != 0:
mean, stDev, median = getMeanAndStDevError(totalErrors)
infoA = "Total mean: " "," + str(mean), ", st dev: " + str(stDev) + " ALPHA: " + str(ALPHA)
logger.info(infoA)
return mean, stDev, totalErrors
if not headerLine == None:
outputFileHandle.write(headerLine)
output = ""
for element in inputList:
output = output + str(element) + "\t"
output = output.strip()
output = output + '\n'
outputFileHandle.write(output)
outputFileHandle.close()
if __name__ == '__main__':
# findDiffMean(phonemeName)
conv_before = HtkConverter()
conv_before.load(MODEL_URI, HMM_LIST_URI)
# one phoneme
phonemeName = 'A'
hmmModels = [hmm for hmm in conv_before.hmms if hmm.name == phonemeName]
hmmModel1 = hmmModels[0]
# all phonemes
for currHmmModel in conv_before.hmms:
# printModelsMiddleState(currHmmModel, PATH_TO_HMMLIST);
printModelsAllStates(currHmmModel, PATH_TO_HMMLIST)
