def feature_extraction_and_force_alignment(filepath, nstates, phoneHMMs):
"""
handle one .wav file
"""
samples, samplingrate = loadAudio(filepath)
wordTrans = list(path2info(filepath)[2])
phoneTrans = words2phones(wordTrans, prondict)
stateTrans = [phone + '_' + str(stateid) for phone in phoneTrans
for stateid in range(nstates[phone])]
lmfcc_result = mfcc(samples)
mspec_result = mspec(samples)
targets = []
_, viterbi_path = forcedAlignment(lmfcc_result, phoneHMMs, phoneTrans)
targets = [stateTrans[idx] for idx in viterbi_path.astype(np.int16)]
return lmfcc_result, mspec_result, targets
# create a state list
# split a state by the number of states in its HMM.
# E.g. # of states of ah = 3; ah -> ['ah_0', 'ah_1', 'ah_2']
stateList = list()
for ph in phoneHMMs.keys():
for i in range(nstates[ph]):
stateList.append('%s_%d' % (ph, i))
# --------------------------------------------------------------
data = list()
for root, dirs, files in walk(folder_to_extract):
for f in tqdm(files):
if not f.endswith('.wav'):
continue
# do our work
filename = os.path.join(root, f)
sample, srate = loadAudio(filename)
mspec_x = mspec(sample, samplingrate=srate)
lmfcc_x = mfcc(sample, samplingrate=srate)
wordTrans = list(path2info(filename)[2])
phoneTrans = words2phones(wordTrans, prondict)
targets = forcedAlignment(lmfcc_x, phoneHMMs, phoneTrans)
# convert the targets from str to int
idx_targets = [stateList.index(t) for t in targets]
data.append({
'filename': filename,
'lmfcc': lmfcc_x,
'mspec': mspec_x,
'targets': idx_targets
})
kwargs = {data_type: data}
print("check that the intit proba is equal to 1")
print(np.sum(wordTest['startprob']) == 1)
sumTot = 0.0
N = len(wordTest['transmat'])
for i in range(N):
sumTot += np.sum(wordTest['transmat'][i])
#print("les deux nombres suivants doivent etre egaux")
print(sumTot)
print(N)
#print(wordTest['startprob'])
#print(wordTest['transmat'])
###########
filename = 'tidigits/disc_4.1.1/tidigits/train/man/nw/z43a.wav'
samples, samplingrate = lab3_tools.loadAudio(filename)
lmfcc = lab1.mfcc(samples)
wordTrans = list(lab3_tools.path2info(filename)[2])
print(wordTrans)
#should be ['z', '4', '3']
phoneTrans = words2phones(wordTrans, prondict, addShortPause=True)
print(phoneTrans)
#should be ['sil', 'z', 'iy', 'r', 'ow', 'f', 'ao', 'r', 'th', 'r', 'iy', 'sil']
stateTrans = [
phone + '_' + str(stateid) for phone in phoneTrans
for stateid in range(nstates[phone])
]
print(stateTrans)
"""
from lab1_proto import mfcc
from lab3_tools import loadAudio
from lab3_tools import path2info
from lab3_proto import words2phones
from prondict import prondict
from lab2_proto import concatHMMs
from lab3_tools import frames2trans
import numpy as np
# ==============
from lab3_proto import forcedAlignment
if __name__ == "__main__":
filename = 'tidigits/disc_4.1.1/tidigits/train/man/nw/z43a.wav'
# Get the mfcc feature vectors
samples, samplingrate = loadAudio(filename)
lmfcc = mfcc(samples)
# =================================================
# Get the word level transcription
wordTrans = list(path2info(filename)[2])
# get the phone level transcription
phoneTrans = words2phones(wordTrans, prondict)
# print(phoneTrans)
# ==================================================================
# combine the HMMs according to the phone level transcription
phoneHMMs = np.load('lab2_models_all.npz',allow_pickle=True)['phoneHMMs'].item()
utteranceHMM = concatHMMs(phoneHMMs, phoneTrans)
# translate the state (idx) to a name
phones = sorted(phoneHMMs.keys())
nstates = dict()
for ph in phones: