def get_timestamps(train_files, test_files, output_folder):
for idx, files in enumerate([train_files, test_files]):
pairs = []
for i in range(len(files)):
for j in range(i + 1, len(files)):
file1 = files[i].split('/')[-1].split('.')[0]
file2 = files[j].split('/')[-1].split('.')[0]
if file1 == file2:
sorter = file1.split('-')
s = "-"
num_zeros = 10 - len(sorter[1])
for _ in range(num_zeros):
s += '0'
sorter = sorter[0] + s + sorter[1]
if files[i].endswith('.txt'):
pairs.append((files[j], files[i], sorter))
else:
pairs.append((files[i], files[j], sorter))
pairs = sorted(pairs, key=lambda x: x[-1])
for wav, txt, _ in pairs:
try:
original = []
with open(txt, 'r', encoding='utf8') as f:
for line in f:
original.append(line.strip())
phoneme_alignments, word_alignments = align.align(wav, txt)
output_file = txt.split('/')[-1]
if idx == 0:
output_path = os.path.join(output_folder, 'train')
if idx == 1:
output_path = os.path.join(output_folder, 'test')
if not os.path.exists(output_path):
os.makedirs(output_path)
with open(os.path.join(output_path, output_file), 'w', encoding='utf8') as f:
for word, start, stop in word_alignments:
f.write(word + ',' + str(start) + ',' + str(stop) + '\n')
print('Alignment successful', wav, txt)
except:
print('Alignment failed: ', wav, txt)
print('-----------------------------')
print()
for names in items:
if names.endswith(".txt"):
txtfilelist.append(names)
# Iterate through text files, performing aligner and saving Praat as we go
# for file in txtfilelist:
for ff in range(0,len(txtfilelist)):
curr_text = txtfilelist[ff]
curr_wav = curr_text.replace('txt', "wav")
curr_TextGird = curr_text.replace('txt', "TextGrid")
print([str(ff)+': '+curr_text])
phoneme_alignments, word_alignments = align.align(wavdirname+'/'+curr_wav, txtdirname+'/'+curr_text,TGdirname+'/'+curr_TextGird)
with open(Path(matdirname,curr_text), 'w') as fp:
fp.writelines('Phonemes\n')
for pp in range(1,len(phoneme_alignments)):
fp.write(' '.join('%s' % x for x in phoneme_alignments[pp]))
fp.writelines('\n')
fp.writelines('Words\n')
for pp in range(1,len(word_alignments)):
fp.write(' '.join('%s' % x for x in word_alignments[pp]))
fp.writelines('\n')
## For Spanish (unfortunately P2FA doesn't have Spanish support)
txtdirname = r'/home/sakkol/Documents/Forced_Alignment/FORCE/Fast-Spanish/txt'
def test_aligner(self):
align.align(self.input_wav, self.input_transcription, self.outfile)
self.assertTrue(filecmp.cmp(self.outfile, self.true_alignment_file))
def extract_phoneme_data(args):
audio_file_name, lyrics_file_name, audio_length = args
print(audio_file_name)
# Extract phonemes using Penn's force aligner
ph_align, w_align = align.align(audio_file_name, lyrics_file_name)
step = params.frame_period / 1000
phoneme_position = 0
phoneme_array = []
x = 0
while x < audio_length:
if ph_align[phoneme_position][1] <= x * step:
if ph_align[phoneme_position][2] > x * step:
phoneme_array.append(ph_align[phoneme_position][0])
x = x + 1
elif phoneme_position + 1 < len(ph_align):
phoneme_position = phoneme_position + 1
else:
x = x + 1
phoneme_array.append(ph_align[phoneme_position][0])
else:
phoneme_array.append("sp")
x = x + 1
x = 0
timing_array = []
while x < audio_length:
phoneme = phoneme_array[x]
counter = 0
while x + counter < audio_length and phoneme_array[x +
counter] == phoneme:
counter = counter + 1
numerator = 1
for y in range(counter):
# timing_array.append(numerator / counter)
timing = numerator / counter
if timing <= 0.333:
timing_array.append(0)
elif timing <= 0.666:
timing_array.append(1)
else:
timing_array.append(2)
numerator = numerator + 1
x = x + counter
pre_phoneme_array = []
post_phoneme_array = []
for y in range(audio_length):
try:
pre_phoneme_array.append(phoneme_array[y - 1])
except IndexError:
pre_phoneme_array.append("sp")
try:
post_phoneme_array.append(phoneme_array[y + 1])
except IndexError:
post_phoneme_array.append("sp")
phoneme_position_data = pd.DataFrame(phoneme_array, columns=['Phoneme'])
pre_phoneme_position_data = pd.DataFrame(pre_phoneme_array,
columns=['Pre_phoneme'])
post_phoneme_position_data = pd.DataFrame(post_phoneme_array,
columns=['Post_phoneme'])
phoneme_timing_data = pd.DataFrame(timing_array,
columns=['Phoneme_timings'])
phoneme_data = pd.concat([
phoneme_position_data, phoneme_timing_data, pre_phoneme_position_data,
post_phoneme_position_data
],
axis=1)
return phoneme_data