def save_intervalframe_to_textgrid(framedict, filepath, encoding='utf-8'):
"""Write a dict of IntervalFrames in a textgrid-File.
Arguments:
framedict -- Dictionary of dataframes. The keys become tier
names in the textgrid file
filepath -- Path + filename of the file to be written.
Keyword arguments:
encoding: character encoding to save textgrid file
"""
if len(framedict) < 1:
print "invalid data!"
return
mytextgrid = tgt.TextGrid()
for tier_name in framedict.keys():
newtier = framedict[tier_name]
if len(newtier.columns) == 3:
mytier = tgt.IntervalTier(name=tier_name)
for row in newtier.index:
myinterval = tgt.Interval(newtier[newtier.columns[0]][row],
newtier[newtier.columns[1]][row],
newtier[newtier.columns[2]][row])
mytier.add_interval(myinterval)
elif len(newtier.columns) == 2:
mytier = tgt.PointTier(name=tier_name)
for row in newtier.index:
mypoint = tgt.Point(newtier[newtier.columns[0]][row],
newtier[newtier.columns[1]][row])
mytier.add_point(mypoint)
mytextgrid.add_tier(mytier)
tgt.write_to_file(mytextgrid, filepath, encoding=encoding, format="long")
def save_annotations(self, filename, tiers=['cycles', 'holds'],
filetype='textgrid', merge_holds=False):
"""Save annotations to file."""
if filetype not in ['textgrid', 'eaf', 'table']:
raise ValueError('Unsupported file type: {}'.format(filetype))
tg = tgt.TextGrid()
if 'holds' in tiers or merge_holds:
# holds = tgt.IntervalTier(name='holds')
# for start, end in self.holds:
# holds.add_interval(tgt.Interval(start, end, 'hold'))
if not merge_holds:
tg.add_tier(self.holds)
if 'cycles' in tiers:
if merge_holds:
segments_merged= tgt.IntervalTier(name='cycles')
tg.add_tier(self.merge_holds(self.segments, self.holds))
else:
tg.add_tier(self.segments)
if len(tg.tiers):
filetype = 'short' if filetype == 'textgrid' else filetype
tgt.write_to_file(tg, filename, format=filetype)
def main():
# Parse the command-line arguments.
args = parse_arguments()
tg_path = args['tg_path']
offset_start = args['offset_start']
offset_end = args['offset_end']
outpath = args['outpath']
# Read the TextGrid
tg = tgt.read_textgrid(tg_path)
tg_part = tgt.TextGrid()
if offset_start is None and offset_end is None:
raise Exception(
'At least one of offset_start and offset_end must be specified.')
elif offset_start is None:
offset_start = tg.start_time
elif offset_end is None:
offset_end = tg.end_time
for tr in tg:
intr_part = tr.get_annotations_between_timepoints(
offset_start, offset_end)
tier_part = tgt.IntervalTier(name=tr.name,
start_time=tr.start_time,
end_time=tr.end_time,
objects=intr_part)
tg_part.add_tier(tier_part)
if outpath is None:
tg_dirname, tg_filename = os.path.split(tg_path)
outpath = os.path.splitext(tg_filename)[0] + '_part.TextGrid'
tgt.write_to_file(tg_part, outpath)
def get_textgrid(self):
tg = tgt.TextGrid()
t = tgt.IntervalTier(name='Word')
for w in self.words.segments:
t.add_interval(tgt.Interval(w.start, w.end, w.text))
tg.add_tier(t)
t = tgt.IntervalTier(name='Phoneme')
for ph in self.phonemes.segments:
t.add_interval(tgt.Interval(ph.start, ph.end, ph.text))
tg.add_tier(t)
return tgt.io.export_to_long_textgrid(tg)
def prepare_textgrid(self, df, offset):
grid = tgt.TextGrid()
tier = tgt.IntervalTier()
tier.name = "Context"
grid.add_tier(tier)
for x in df.index:
start = df.loc[x]["coq_word_starttime_1"]
end = df.loc[x]["coq_word_endtime_1"]
text = df.loc[x]["coq_word_label_1"]
interval = tgt.Interval(start - offset, end - offset)
interval.text = text
tier.add_interval(interval)
return grid
def init_textgrid(infile, duration, *tiers):
tg = tgt.TextGrid()
if infile is not None:
logging.info("reading TextGrid %s" % infile)
tg = tgt.io.read_textgrid(infile)
result = [tg]
for tier in tiers:
if tg.has_tier(tier):
logging.info("overwriting tier %s" % tier)
tg.delete_tier(tier)
tier = tgt.IntervalTier(name=tier, start_time=0, end_time=duration)
result.append(tier)
return tuple(result)
def main():
ap = argparse.ArgumentParser()
ap.add_argument(
'shift',
help='offset by which to shift the boundaries (positive or negative)',
type=float)
ap.add_argument('file', help='the textgrid file', type=str)
ap.add_argument('-e',
'--encoding',
help='file encoding (default "utf-8")',
default='utf-8',
type=str)
ap.add_argument('-f',
'--format',
help='the output format (default "short")',
default='short',
type=str)
ap.add_argument(
'-o',
'--outfile',
help='the output file (defaults to inputfile.shifted.Extension)',
type=str)
arguments = ap.parse_args()
# Read file
try:
tg = tgt.read_textgrid(filename=arguments.file,
encoding=arguments.encoding)
except IOError:
print('An error occurred reading file {file}'.format(
file=arguments.file))
sys.exit(1)
# Create new textgrid
if arguments.outfile is None:
basename, extension = os.path.splitext(arguments.file)
output_filename = basename + '.shifted' + extension
else:
output_filename = arguments.outfile
tg_shifted = tgt.TextGrid(filename=output_filename)
# Shift boundaries
for tier in tg:
ts = tgt.util.shift_boundaries(tier, arguments.shift, 0)
tg_shifted.add_tier(ts)
# Write file
tgt.write_to_file(textgrid=tg_shifted,
filename=tg_shifted.filename,
format=arguments.format,
encoding=arguments.encoding)
def writeTextGrid(outfile, word_alignments):
tg = tgt.TextGrid()
phone_tier = tgt.IntervalTier(name='phone')
word_tier = tgt.IntervalTier(name='word')
for data in word_alignments:
word = data[0]
phones = data[1:]
if len(phones) > 0:
start_time = phones[0][1]
end_time = phones[-1][2]
word_tier.add_interval(
tgt.Interval(start_time, end_time, text=word))
for (p, p_start, p_end) in phones:
phone_tier.add_interval(tgt.Interval(p_start, p_end, text=p))
tg.add_tier(phone_tier)
tg.add_tier(word_tier)
tgt.io.write_to_file(tg, outfile, format='long')
def format(self, syncmap):
try:
import tgt
except ImportError as exc:
self.log_exc(u"Python module tgt is not installed", exc, True, ImportError)
# from https://github.com/hbuschme/TextGridTools/blob/master/tgt/io.py
textgrid = tgt.TextGrid()
tier = tgt.IntervalTier(name="Token")
for fragment in syncmap.fragments:
begin = float(fragment.begin)
end = float(fragment.end)
text = fragment.text_fragment.text
if text == u"":
text = u"SIL"
interval = tgt.Interval(begin, end, text=text)
tier.add_interval(interval)
textgrid.add_tier(tier)
if self.variant == self.DEFAULT:
msg = tgt.io.export_to_long_textgrid(textgrid)
else:
msg = tgt.io.export_to_short_textgrid(textgrid)
return gf.safe_unicode(msg)
def stitch_textgrid(batch_title, sequenced_title, input2b_path, input2_path,
output3_path):
combined_intervals = []
new_tg = tgt.TextGrid()
new_phone_tier = tgt.IntervalTier()
final_phone_tier = tgt.IntervalTier()
new_word_tier = tgt.IntervalTier()
last_dur = 0.0
for i, title in enumerate(sequenced_title):
wave_file = wave.open(os.path.join(input2b_path, title + '.wav'), 'rb')
frameRate = wave_file.getframerate()
n_frames = wave_file.getnframes()
dur = n_frames / frameRate
f0_start_time = 0.0
f0_end_time = dur
tg = tgt.read_textgrid(os.path.join(input2_path, title + '.TextGrid'))
# Load name of all tiers
tier_names = tg.get_tier_names()
words_tier_name = [name for name in tier_names if 'words' in name][0]
words_tier = tg.get_tier_by_name(words_tier_name)
phones_tier_name = [name for name in tier_names if 'phones' in name][0]
phones_tier = tg.get_tier_by_name(phones_tier_name)
word_annotations = words_tier.get_annotations_between_timepoints(
f0_start_time, f0_end_time)
phone_annotations = phones_tier.get_annotations_between_timepoints(
f0_start_time, f0_end_time)
word_intervals = []
for interval in word_annotations:
interval.end_time = interval.end_time + last_dur
interval.start_time = interval.start_time + last_dur
word_intervals.append(interval)
if word_intervals[-1].end_time > last_dur + f0_end_time:
word_intervals[-1].end_time = last_dur + f0_end_time
phone_intervals = []
for j, interval in enumerate(phone_annotations):
interval.end_time = interval.end_time + last_dur
interval.start_time = interval.start_time + last_dur
if interval.text != 'sil' and interval.text != 'sp':
phone_intervals.append(interval)
elif i == len(sequenced_title) - 1 and j == len(
phone_annotations) - 1:
phone_intervals.append(interval)
if phone_intervals[-1].end_time > last_dur + f0_end_time:
phone_intervals[-1].end_time = last_dur + f0_end_time
new_word_tier.add_annotations(word_intervals)
new_phone_tier.add_annotations(phone_intervals)
last_dur += dur
phones_tier_copy = new_phone_tier.get_copy_with_gaps_filled(
start_time=None, end_time=None, empty_string='')
# Replace all sil and sp intervals with <sil> tag
#store these intervals to a list so that we can add them to the other tiers
sil_intervals = []
phone_intervals = []
for interval in phones_tier_copy:
if interval.text == '':
interval.text = 'sil'
sil_intervals.append(interval)
else:
phone_intervals.append(interval)
final_phone_tier.add_annotations(phone_intervals)
final_phone_tier.add_annotations(sil_intervals)
final_phone_tier.name = phones_tier_name
new_word_tier.name = words_tier_name
new_tg.add_tier(new_word_tier)
new_tg.add_tier(final_phone_tier)
tgt.write_to_file(new_tg,
os.path.join(output3_path, batch_title + '.TextGrid'),
format='short')
def prepare_textgrids(self,
order=None,
one_grid_per_match=False,
remember_time=False):
"""
Parameters
----------
order: list
A list of columns that specifies the order of the text grid tiers.
"""
self.feature_timing = dict()
grids = {}
if "coquery_invisible_origin_id" not in self.df.columns:
one_grid_per_match = True
if one_grid_per_match:
key_columns = [self.resource.file_name, self.resource.corpus_id]
else:
key_columns = [self.resource.file_name]
for i in self.file_data.index:
grid_id = tuple(self.file_data.iloc[i][key_columns])
grids[grid_id] = tgt.TextGrid()
if ("corpus_starttime" in options.cfg.selected_features
and "corpus_endtime" in options.cfg.selected_features):
self.feature_timing["corpus_id"] = ("corpus_starttime",
"corpus_endtime")
if order:
features = []
for column in order:
if column.startswith("coq_"):
name = column.rpartition("coq_")[-1].rpartition("_")[0]
else:
name = column
if (name not in features
and not name.startswith("coquery_invisible")):
features.append(name)
else:
features = options.cfg.selected_features
tiers = set([])
for rc_feature in [
x for x in features
if (not x.startswith(("func_", "coquery_", "db_")))
]:
hashed, tab, feature = (
self.resource.split_resource_feature(rc_feature))
if tab == "segment":
# the segment table is hard-wired:
start_label = "{}_starttime".format(tab)
end_label = "{}_endtime".format(tab)
self.feature_timing[rc_feature] = (start_label, end_label)
else:
# determine the table that contains timing information by
# following the table path:
self.resource.lexicon.joined_tables = ["corpus"]
self.resource.lexicon.table_list = ["corpus"]
self.resource.lexicon.add_table_path("corpus_id",
"{}_id".format(tab))
for current_tab in self.resource.lexicon.joined_tables:
# check if timing information has been selected for the
# current table from the table path:
start_label = "{}_starttime".format(current_tab)
end_label = "{}_endtime".format(current_tab)
# if so, set the timing entry for the current feature
# to these timings:
if (start_label in options.cfg.selected_features
and end_label in options.cfg.selected_features
) and not (rc_feature.endswith(
("endtime", "starttime"))):
self.feature_timing[rc_feature] = (start_label,
end_label)
rc_feat = "{}_{}".format(tab, feature)
if hashed is not None:
link, res = get_by_hash(hashed)
tier_name = "{}.{}_{}".format(res.db_name, link.rc_to)
else:
tier_name = rc_feat
if (rc_feat not in [start_label, end_label]
and tier_name not in tiers):
# ... but only if it is not containing timing information
for x in grids:
grids[x].add_tier(tgt.IntervalTier(name=tier_name))
tiers.add(tier_name)
for col in [
x for x in features if x.startswith(("func", "coquery", "db"))
]:
# FIXME:
# db and func columns are never treated as lexicalized columns.
# The fix for this is probably not quite trivial.
tier_name = self.session.translate_header(col)
for x in grids:
grids[x].add_tier(tgt.IntervalTier(name=tier_name))
tiers.add(tier_name)
# if there is no tier in the grids, but the corpus times were
# selected, add a tier for the corpus IDs in all grids:
if (not (grids[list(grids.keys())[0]].tiers)
and ("corpus_starttime" in options.cfg.selected_features
and "corpus_endtime" in options.cfg.selected_features)
and not self._artificial_corpus_id):
self._artificial_corpus_id = True
for f in grids:
grids[f].add_tier(tgt.IntervalTier(name="corpus_id"))
if remember_time:
for f in grids:
grids[f].add_tier(tgt.PointTier(name="Original timing"))
return grids
# Usage: python segment_laughter.py <input_audio_file> <stored_model_path> <output_folder> <save_to_textgrid>
if __name__ == '__main__':
if parse_inputs():
input_path, model_path, output_path, threshold, min_length, save_to_textgrid = parse_inputs(
)
min_length = seconds_to_frames(min_length)
laughs = laugh_segmenter.segment_laughs(input_path, model_path,
output_path, threshold,
min_length, save_to_textgrid)
print("found %d laughs." % (len(laughs)))
if not save_to_textgrid:
for laugh in laughs:
print(laugh)
else:
tg = tgt.TextGrid()
laughs_tier = tgt.IntervalTier(
name='laughter',
objects=[
tgt.Interval(l['start'], l['end'], 'laugh') for l in laughs
])
tg.add_tier(laughs_tier)
fname = os.path.splitext(os.path.basename(input_path))[0]
tgt.write_to_file(
tg, os.path.join(output_path, fname + '_laughter.TextGrid'))
print('Saved laughter segments in {}'.format(
os.path.join(output_path, fname + '_laughter.TextGrid')))
import tgt grid = tgt.TextGrid(filename="test") tier = tgt.IntervalTier(start_time=0, end_time=5, name="mot") label = tgt.core.Interval(2, 3, "word") tier.add_annotation(label) grid.add_tier(tier) tgt.write_to_file(grid, "/home/leferrae/Desktop/These/test.textgrid")
