def _textgrid2sfs(txtlines, output_path):
logger = logging.getLogger('mtts')
textgrid_path = os.path.join(output_path, 'textgrid/mandarin_voice')
sfs_path = os.path.join(output_path, 'sfs')
csv_path = os.path.join(output_path, 'csv')
os.system('mkdir -p %s' % sfs_path)
os.system('mkdir -p %s' % csv_path)
for line in txtlines:
numstr, txt = line.split(' ', 1)
textgrid_file = os.path.join(textgrid_path, numstr + '.TextGrid')
csv_file = os.path.join(csv_path, numstr + '.csv')
sfs_file = os.path.join(sfs_path, numstr + '.sfs')
if os.path.exists(textgrid_file):
# textgrid to csv
tgrid = tg.read_textgrid(textgrid_file)
tg.write_csv(tgrid, csv_file, sep=' ', header=False, meta=False)
# csv to sfs
total_list = []
with open(csv_file) as fid:
for line in fid.readlines():
#start, end, name, label = line.strip().split(' ')
csv_list = line.strip().split(' ')
if csv_list[3] == 'phones':
total_list.append(_standard_sfs(csv_list))
with open(sfs_file, 'w') as fid:
for item in total_list:
fid.write(' '.join(item) + '\n')
else:
logger.warning('--Miss: %s' % textgrid_file)
with open(os.path.join(output_path, 'error.log'), 'a+') as fid:
fid.write('--Miss: %s \n' % textgrid_file)
def analyze_sentiment():
input_file, language, output_file = parse_arguments()
# Import pattern module corresponding to language
pattern = importlib.import_module(f'pattern.{language}')
# Load file
blob = open(input_file, 'rb').read()
m = magic.Magic(mime_encoding=True)
encoding = m.from_buffer(blob)
tgrid = textgrid.read_textgrid(input_file, encoding)
df = pd.DataFrame(tgrid)
# Filter for Transcripts, remove empty
df = df[df['tier'] == 'Transcript']
df = df[df['name'] != '']
# Calculate sentiment
# TODO: figure out how to unpack tuples to multiple columns:
df['sentiment_polarity'] = df.apply(lambda row: pattern.sentiment(row['name'])[0], axis=1)
df['sentiment_subjectivity'] = df.apply(lambda row: pattern.sentiment(row['name'])[1], axis=1)
# Save final result
print(df.head())
df.to_csv(output_file, index=False)
def test_read_short():
tgrid = textgrid.read_textgrid(example_file2)
assert len(tgrid) == 1
entry = tgrid[0]
assert entry.start == 0
assert entry.stop == 5.537098932314087
assert entry.name == "z"
assert entry.tier == "Mary"
def test_read_long():
tgrid = textgrid.read_textgrid(example_file1)
assert len(tgrid) == 11
entry = tgrid[0]
assert entry.start == 0
assert entry.stop == 5.537098932314087
assert entry.name == ""
entry2 = tgrid[-1]
assert entry2.name == "cool"
assert entry2.tier == "bell"
def parse_tg(tgd, ad, uttd, labd):
problemnums = [
".1", ".2", ".3", ".4", ".5", ".6", ".7", ".8", ".9", "1.0", "2.0",
"3.0", "4.0", "5.0", "6.0", "7.0", "8.0", "9.0"
]
# ------------------------------ CHANGE BASED ON TIER NAMES FOR SPEAKERS ----------------------
speakers = ["one", "two"]
counter = 0.101
for file in sorted(os.listdir(tgd)):
if file.endswith(".TextGrid"):
print("Working on: ", file)
# ----------------------------- CHANGE BASED ON FILE NAME ---------------------------
dyad = file.split("C0")[1].split("_")[0]
tg = textgrid.read_textgrid(os.path.join(tgd, file))
df = pd.DataFrame(tg)
for speaker in speakers:
print(" Speaker: ", speaker)
# ----------------------------- CHANGE BASED ON DYAD vs. MONOLOGUE ---------------------------
audio_file = os.path.join(ad, dyad + "_" + speaker + ".wav")
# data = df.loc[df["tier"] == (speaker + " [main]")]
data = df.loc[df["tier"] == speaker]
for index, row in data.iterrows():
# row[name] refers to the text of the utterance
if row["name"] != "":
write_audio(audio_file, row['start'], row['stop'],
counter, dyad, speaker, uttd)
write_lab(row["name"], counter, dyad, speaker, labd)
counter = round(counter + .001, 3)
for p in problemnums:
if str(counter) == p:
counter = round(counter + .001, 3)
break
while abs(
decimal.Decimal(
str(counter)).as_tuple().exponent) < 3:
counter = round(counter + .001, 3)
def extract_features(dataframe):
input = []
target = []
segment_labels = []
for i, (file, emotion) in enumerate(dataframe.values):
script_path = '/scratch/speech/modularProsodyTagger/mod01.praat'
index = file.rfind('/')
basename = file[(index + 1):-4]
directory = file[:60] + basename[:-5] + '/'
endpoint = '/scratch/speech/textgrids/'
cmd = 'praat --run {} {} {} {}'.format(script_path, directory,
basename, endpoint)
#os.system(cmd)
tgrid_path = endpoint + basename + '_result.TextGrid'
tgrid = textgrid.read_textgrid(tgrid_path)
tgrid_df = pd.DataFrame(tgrid)
data = []
indices = []
labels = []
sample_rate, data = wavfile.read(file)
for start, stop, name, tier in tgrid_df.values:
if tier != 'silences':
break
else:
indices.append(round(stop * sample_rate))
labels.append(name)
#data = data.tolist()
data = [data[i:j] for i, j in zip(([0] + indices)[:-1], indices)]
#print(data)
input.append(data)
target.append(encode[emotion])
segment_labels.append(labels)
print(i)
return input, target, segment_labels
def load_silences(silences_file, time_series):
# Load file
blob = open(silences_file, 'rb').read()
m = magic.Magic(mime_encoding=True)
encoding = m.from_buffer(blob)
silences_tgrid = textgrid.read_textgrid(silences_file, encoding)
silences = pd.DataFrame(silences_tgrid)
# Add column
def isSilent(timestamp):
filt = (silences['start'] <= timestamp) & (timestamp <
silences['stop'])
silence = silences.loc[filt, 'name']
if len(silence) > 0:
return silence.iat[0] == 'silent'
return True
time_series['silence'] = time_series.apply(
lambda row: isSilent(row.timestamp), axis=1)
def link_uttid_start(tgd):
problemnums = [
".1", ".2", ".3", ".4", ".5", ".6", ".7", ".8", ".9", "1.0", "2.0",
"3.0", "4.0", "5.0", "6.0", "7.0", "8.0", "9.0"
]
# ------------------------------ CHANGE BASED ON TIER NAMES FOR SPEAKERS ----------------------
speakers = ["one", "two"]
for file in sorted(os.listdir(tgd)):
counter = 0.101
if file.endswith(".TextGrid"):
print("Working on: ", file)
# next two lines take file name and pull out the identifier for the dyad / monologue speaker
# ----------------------------- CHANGE BASED ON FILE NAME ---------------------------
dyad = file.split("C0")[1].split("_")[0]
tg = textgrid.read_textgrid(os.path.join(tgd, file))
df = pd.DataFrame(tg)
for speaker in speakers:
print(" Speaker: ", speaker)
data = df.loc[df["tier"] == speaker]
for index, row in data.iterrows():
if row["name"] != "":
#write_start_to_uttID(row['start'], row['stop'], counter, dyad, speaker)
write_len(row["name"], counter, dyad, speaker,
row['start'], row['stop'])
counter = round(counter + .001, 3)
for p in problemnums:
if str(counter) == p:
counter = round(counter + .001, 3)
break
while abs(
decimal.Decimal(
str(counter)).as_tuple().exponent) < 3:
counter = round(counter + .001, 3)
import textgrid as tg
tgrid = tg.read_textgrid(r'E:\Biaobei_Demo\000001.TextGrid', 'Phon')
for entry in tgrid:
print(entry.name)
print(len(tgrid))
print('Done!')
@License : (C)Copyright 2019-2020
@Desc : stats some complex infos in TextGrids
"""
import textgrid as tg
import os
input_textgrid_dir = r'E:\006_TTS\biaobei\CMU_Demo'
save_result_file = r'E:\006_TTS\biaobei\CMU_Demo_complex.txt'
results = {}
for file_name in os.listdir(input_textgrid_dir):
if file_name.find('.TextGrid') != -1:
print(file_name)
input_textgrid_file = os.path.join(input_textgrid_dir, file_name)
ph_tiers = tg.read_textgrid(input_textgrid_file, 'phones')
wd_tiers = tg.read_textgrid(input_textgrid_file, 'words')
for wd_tier in wd_tiers:
cur_word = wd_tier.name
for punc in [',', '，', '。', '.']:
cur_word = cur_word.replace(punc, '')
cur_phons = []
for ph_tier in ph_tiers:
if ph_tier.start >= wd_tier.start and ph_tier.stop <= wd_tier.stop:
cur_phons.append(ph_tier.name)
if cur_word not in results:
results[cur_word] = [' '.join(cur_phons)]
else:
if ' '.join(cur_phons) not in results[cur_word]:
results[cur_word].append(' '.join(cur_phons))
with open(save_result_file, 'w', encoding='utf-8') as wid:
def test_points():
tgrid = textgrid.read_textgrid(example_file3)
assert len(tgrid) == 2