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 parse_grid(grid_path):
'''
note : parse a textgrid and then split the wav matches the textgrid
arg :
grid_path : textgrid path to parse
'''
dir_path = os.path.dirname(grid_path)
grid_filename = os.path.basename(grid_path)
name = os.path.splitext(grid_filename)[0]
wav_path = os.path.join(dir_path, name + ".wav")
target_tier = "comma"
tg_obj = tgt.read_textgrid(grid_path)
# get objects(textgrid-tier & wav) matches the grid path
tier_obj = tg_obj.get_tier_by_name(target_tier)
wav_obj, sr = librosa.load(wav_path, sr=None)
for idx in range(len(tier_obj)):
part = tier_obj[idx]
time_s = librosa.time_to_samples(part.start_time, sr)
time_e = librosa.time_to_samples(part.end_time, sr)
librosa.output.write_wav('{}_{}.wav'.format(name, idx),
wav_obj[time_s:time_e], sr)
with open("{}_{}.txt".format(name, idx), "w") as f:
f.write(part.text)
def add_lemmas(title, input1_path, output_path):
# Load textgrid
tg = tgt.read_textgrid(os.path.join(input1_path, title + '.TextGrid'))
tier_names = tg.get_tier_names()
# Load pos tier
pos_tier_name = [name for name in tier_names if 'pos' in name][0]
pos_tier = tg.get_tier_by_name(pos_tier_name)
# Load words tier
words_tier_name = [name for name in tier_names if 'words' in name][0]
words_tier = tg.get_tier_by_name(words_tier_name)
# Start empty lemmas tier
lemmas_tier = tgt.IntervalTier()
lemmas_tier_name = [name for name in tier_names
if 'words' in name][0].replace('words', 'lemmas')
lemmas_tier.name = lemmas_tier_name
# Generate lemma intervals
lemmas_intervals = [
tgt.Interval(w_interval.start_time, w_interval.end_time,
lemmatize_word(w_interval.text, pos_tier[i].text))
for i, w_interval in enumerate(words_tier)
]
# Add lemmas to tier
lemmas_tier.add_annotations(lemmas_intervals)
tg.add_tier(lemmas_tier)
tgt.write_to_file(tg,
os.path.join(output_path, title + '.TextGrid'),
format='short')
def get_textgrid_sa(mfa_file, merge_shorter=0.15, pause_tokens=[""]):
read_textgrid = tgt.read_textgrid(mfa_file, include_empty_intervals=False)
[words, start_time, end_time] = read_word_alignment(read_textgrid)
assert len(words) == len(start_time) == len(end_time)
stack = []
out_words = []
for i in range(len(words)):
if words[i] in pause_tokens: # pause skip
continue
if stack:
if start_time[i] - stack[-1][-1] > merge_shorter:
# determine how much long is the pause
out_words.append([words[i]])
stack.append([start_time[i], end_time[i]])
else:
stack[-1][-1] = end_time[i]
out_words[-1].append(words[i])
else:
stack.append([start_time[i], end_time[i]])
out_words.append([words[i]])
return stack, out_words
def generate_punctuation_dictionary(titles, input_path, output_path):
punctuation_dict = {}
# Gather bp intervals first
for title in titles:
# Load the textgrid
tg = tgt.read_textgrid(os.path.join(input_path, title + '.TextGrid'))
# Load name of all tiers
tier_names = tg.get_tier_names()
# Select a tier whose name contains 'bp'
bp_tier_name = [name for name in tier_names if 'bp' in name][0]
bp_tier = tg.get_tier_by_name(bp_tier_name)
# Tally up all the pos in the textgrids
for interval in bp_tier:
if interval.text not in punctuation_dict.keys():
punctuation_dict[interval.text] = 1
else:
punctuation_dict[interval.text] += 1
# Select a tier whose name contains 'fp'
fp_tier_name = [name for name in tier_names if 'fp' in name][0]
fp_tier = tg.get_tier_by_name(fp_tier_name)
# Tally up all the pos in the textgrids
for interval in fp_tier:
if interval.text not in punctuation_dict.keys():
punctuation_dict[interval.text] = 1
else:
punctuation_dict[interval.text] += 1
punct_tuples = sorted(punctuation_dict.items(),
key=lambda x: x[1],
reverse=True)
# 0.8 means that of all the frequencies we keep the highest 80% (for stimuli this was 100%)
# excluding the lowest frequency words might help make the model more robust
punct_freqs = sorted(set([punct_tuple[1] for punct_tuple in punct_tuples]),
reverse=True)
punct_freqs = punct_freqs[:math.floor(len(punct_freqs) * 0.8)]
punct_list = [
punct_tuple[0] for punct_tuple in punct_tuples
if punct_tuple[1] in punct_freqs
]
# Extract the pos found in the corpus add the unk tag
punct_list = ['<unk>'] + punct_list
# Convert the list into hot vector dictionary
hv_dict = {}
for j, v in enumerate(punct_list):
hv_dict[v] = [int(i) for i in np.eye(len(punct_list), dtype=int)[j]]
with open(os.path.join(output_path, 'punctuation_dictionary.json'),
'w') as f:
json.dump(hv_dict, f)
def segment_phonemes(whichFold):
'''
for each phoneme: extract the audio segments from many audio files and concatentate it in a wav file
then one can extract features from that audio (see extractFeatures script)
all_audio_seg: the concatenated audio
'''
annoFiles = get_list_anno_files(AUDIO_REC_IDS, whichFold)
for annoFile in annoFiles: # loop in all annotation files
all_audio_seg = [np.empty([0, 0], dtype='int16')] * len(
PHONEMELIST
) # chunks of concatenated curr_file_audio for each phonemes
audioFile = annoFile.replace('.TextGrid', '.wav')
fs, curr_file_audio = scipy.io.wavfile.read(
os.path.join(audioPath, audioFile))
# get 1 channel
if np.shape(curr_file_audio)[-1] == 2:
curr_file_audio = curr_file_audio[:, 0]
print audioFile, fs, curr_file_audio.shape
######### read phoneme annotations
tgfile = tgt.read_textgrid(os.path.join(annoPath, annoFile))
tgfile.get_tier_names()
tier_phonemes = tgfile.get_tier_by_name("phonemes")
all_audio_seg = concat_phoneme_audio(tier_phonemes, curr_file_audio,
all_audio_seg, fs)
## write the all_audio_seg for each phoneme to file
write_audio_to_file(all_audio_seg)
def _get_label_line(self, path):
'''
Returns aligned labels list where each element corresponds to phone on
audio on i-th timestep. By the paper setting, timestep is 0.01sec
'''
get_textgrid_path = lambda p: p[:-len('flac')] + 'TextGrid'
textgrid_path = get_textgrid_path(path)
textgrid = tgt.read_textgrid(self.root + textgrid_path)
tg_len = textgrid.end_time - textgrid.start_time
labels = [0 for x in range(int(tg_len * 100))]
time_cur = textgrid.start_time
tiers = textgrid.tiers[1]
cur_tier = tiers[0]
cur_tier_i = 0
for i, lab in enumerate(labels):
labels[i] = self.phone2ix[cur_tier.text]
time_cur += 0.01
if cur_tier.end_time < time_cur and cur_tier_i+1 < len(tiers):
cur_tier_i += 1
cur_tier = tiers[cur_tier_i]
return tuple(labels) # tuple is more memory efficient
def add_onsets_rhymes(title, input_path, output_path):
# Load the textgrid
tg = tgt.read_textgrid(os.path.join(input_path, title + '.TextGrid'))
# Load name of all tiers
tier_names = tg.get_tier_names()
# Select a tier whose name contains 'syllables'
sylls_tier_name = [name for name in tier_names if 'sylls' in name][0]
sylls_tier = tg.get_tier_by_name(sylls_tier_name)
# Select a tier whose name contains 'phones'
phones_tier_name = [name for name in tier_names if 'phones' in name][0]
phones_tier = tg.get_tier_by_name(phones_tier_name)
# Start an empty tier for onset-rhymes
onset_rhyme_tier = tgt.IntervalTier()
onset_rhyme_tier_name = [name for name in tier_names
if 'words' in name][0].replace('words', 'OR')
onset_rhyme_tier.name = onset_rhyme_tier_name
onset_rhyme_intervals = []
for syll in sylls_tier._get_annotations():
#print(syll)
phs = phones_tier.get_annotations_between_timepoints(
syll.start_time, syll.end_time)
nucleus_index = calculate_nucleus_index(phs)
# If the first phone contains a number then it means the whole syll has no onset, so we only add a rhyme
if nucleus_index == 0:
onset_rhyme_intervals.append(
tgt.Interval(syll.start_time, syll.end_time, 'R'))
# If the onset is present add onset and rhyme intervals
else:
onset_rhyme_intervals.append(
tgt.Interval(syll.start_time, phs[nucleus_index - 1].end_time,
'O'))
onset_rhyme_intervals.append(
tgt.Interval(phs[nucleus_index].start_time, syll.end_time,
'R'))
# Add all the intervals to the onset rhyme tier
onset_rhyme_tier.add_annotations(onset_rhyme_intervals)
# Add the onset rhyme tier to the TextGrid
tg.add_tier(onset_rhyme_tier)
# Move syll tier after the onset_rhyme_tier
tg.delete_tier(sylls_tier_name)
tg.add_tier(sylls_tier)
tgt.write_to_file(tg,
os.path.join(output_path, title + '.TextGrid'),
format='short')
def read_in_single_textgrid(file_name):
#file_name = os.path.join('/home/zx/Dolphin/Data/textgrid/test/20191011/zcz/1616893', '37429442330632_2019_09_18_01_45_11.TextGrid')
tg_data = tgt.read_textgrid(
file_name) # read a Praat TextGrid file and return a TextGrid object
# print(tg_data)
tier_names = tg_data.get_tier_names() # get names of all tiers
# print (tier_names)
return tg_data
def extract_f0_timepoints(title, input_path, output_path, f0_sr): tg = tgt.read_textgrid(os.path.join(input_path,title+'.TextGrid')) # Original recording's sampling rate audio_sr = 0.005 # Load name of all tiers tier_names = tg.get_tier_names() # Select a tier whose name contains 'sylls' syllables_tier_name = [name for name in tier_names if 'sylls' in name][0] syllables_tier = tg.get_tier_by_name(syllables_tier_name) # In this list we collect the timepoints where we sample the f0 f0_timepoints = [] # Sampling is syllable based, so we loop over each syllable for interval in syllables_tier: # add the first time of the syllable to the list interval_start = interval.start_time f0_timepoints.append(float(interval_start)) # plot the syllable boundary #plt.axvline(x=interval_start, color='orange', linewidth=1) # How many times the default sampling rate approx. fits into this interval n_extractions = round(interval.duration()/f0_sr) # This happens if the interval is shorter than the sampling rate, so we just keep the first value for this interval if n_extractions == 0: pass else: # Based on the N of extractions calculate a new sampling rate, which is specific to this interval sampling_step = interval.duration()/(n_extractions) # Extract based on the new sampling rate # The loop is for n_extractions-1 times because the last point is the first point of the next interval for j in range(0, n_extractions-1): interval_start += sampling_step f0_timepoints.append(interval_start) # Plot extraction points between boundaries #plt.axvline(x=interval_start, color='gray', linewidth=0.5) # Add the very last time of the last syllable interval_start = float(syllables_tier[-1].end_time) f0_timepoints.append(interval_start) # Plot the last point of the last syllable #plt.axvline(x=interval_start, color='orange', linewidth=1) #plt.show() with open(os.path.join(output_path,title+'.json'), 'w') as f: json.dump(f0_timepoints, f)
def get_textgrid_intervals(tgfilepath):
#print(tgfilepath)
try:
tg = tgt.read_textgrid(tgfilepath)
except:
print('Missing textgrid file')
return False
tiers = tg.get_tier_by_name("words")
return tiers.intervals
def child_speech_detector(mom_puzzle_textgrid, v):
""" Detects child speech segments by:
filter out manually annotated mom speech segments
compute child speech band energy per frame, set threhold of being voiced or not
apply median filter to the results of being voiced or not per frame
if two voiced frames are seperated by only 1 min, merge
return time intervals of detected child speech
Args:
mom_puzzle_textgrid: manually annotated mom speech textgrids
v: VoiceActivityDetector object
Returns:
speech_time: the time intervals of detected child speech segments
mom_tier: manually annotated mom speech textgrid tier
"""
speech_time = {}
data = v.data
tg = tgt.read_textgrid(mom_puzzle_textgrid)
mom_tier = tg.get_tier_by_name('Mother')
child_tier = tg.get_tier_by_name('Child')
for i in range(len(mom_tier)):
end_sample = int(round(mom_tier[i].end_time * v.rate))
if i == len(mom_tier) - 1:
start_sample = len(data)
else:
start_sample = int(round(mom_tier[i + 1].start_time * v.rate))
v.data = data[end_sample:start_sample]
detected_windows = v.detect_speech()
for sample_start, speech in detected_windows:
sample_start, speech = int(sample_start), int(speech)
sample_start += end_sample
if speech:
# extend if two intervals speperate by only 1min
region_start_time = sample_start * 1.0 / v.rate
region_start_time = float(
"{0:.2f}".format(region_start_time)) # round to 2 floats
if len(speech_time):
largest_time_prev = max(speech_time.keys())
if region_start_time - 1 <= largest_time_prev:
speech_time[region_start_time] = speech_time[
largest_time_prev]
del speech_time[largest_time_prev]
else:
speech_time[region_start_time] = region_start_time
else:
speech_time[region_start_time] = region_start_time
speech_time = sorted(speech_time.items(), key=operator.itemgetter(1))
return speech_time, mom_tier
def extract_linguistic_labels(title, input1_path, input2_path, output_path):
# Load the f0_timepoints
with open(os.path.join(input1_path, title + '.json')) as f:
f0_timepoints = json.load(f)
# Load textgrid
tg = tgt.read_textgrid(os.path.join(input2_path, title + '.TextGrid'))
tier_names = tg.get_tier_names()
pos_tier_name = [name for name in tier_names if 'pos' in name][0]
# Extract features from TextGrid, except for phones
labels = []
for t in f0_timepoints:
labels_sublist = []
for tier in tier_names:
if 'phones' not in tier and 'words' not in tier:
label = tg.get_tier_by_name(tier).get_annotations_by_time(
t)[0].text
labels_sublist.append(label)
labels.append(labels_sublist)
# Extract syllable boundaries
syllables_tier_name = [name for name in tier_names if 'sylls' in name][0]
syllables_tier = tg.get_tier_by_name(syllables_tier_name)
syllable_boundary_times = [syllables_tier[0].start_time] + [
interval.end_time for interval in syllables_tier
]
syllable_boudaries = [['1'] if t in syllable_boundary_times else ['0']
for t in f0_timepoints]
# Extract word boundaries
wores_tier_name = [name for name in tier_names if 'words' in name][0]
words_tier = tg.get_tier_by_name(wores_tier_name)
word_boundary_times = [words_tier[0].start_time
] + [interval.end_time for interval in words_tier]
word_boudaries = [['1'] if t in word_boundary_times else ['0']
for t in f0_timepoints]
syllable_word_boundaries = [
syllable_boudaries[i] + word_boudaries[i]
for i in range(len(syllable_boudaries))
]
labels = [
syllable_word_boundaries[i] + labels[i] for i in range(len(labels))
]
with open(os.path.join(output_path, title + '.json'), 'w') as f:
json.dump(labels, f)
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 generate_lemmas_dictionary(titles, input1_path, input2_path, output_path):
# Load a control wordlist of the most frequent words in the eng language
wl = load_wordlist(input2_path)
# Collect all of the lemmas from the textgrids in only keep the ones in the control wordlist
# We use the list because we don't want to learn frequent but biased words such as character names or words that are genre-related
lemmas_dict = {}
for title in titles:
# Load the textgrid
tg = tgt.read_textgrid(os.path.join(input1_path, title + '.TextGrid'))
# Load name of all tiers
tier_names = tg.get_tier_names()
# Select a tier whose name contains 'lemmas'
lemmas_tier_name = [name for name in tier_names if 'lemmas' in name][0]
lemmas_tier = tg.get_tier_by_name(lemmas_tier_name)
# Tally up all the lemmas in the textgrids
for interval in lemmas_tier:
if interval.text not in lemmas_dict.keys():
lemmas_dict[interval.text] = 1
else:
lemmas_dict[interval.text] += 1
lemmas_tuples = sorted(lemmas_dict.items(),
key=lambda x: x[1],
reverse=True)
# 0.8 means that of all the frequencies we keep the highest 80% (for stimuli this was 100%)
# excluding the lowest frequency words might help make the model more robust
lemmas_freqs = sorted(set(
[lemmas_tuple[1] for lemmas_tuple in lemmas_tuples]),
reverse=True)
lemmas_freqs = lemmas_freqs[:math.floor(len(lemmas_freqs) * 0.8)]
lemmas_list = [
lemmas_tuple[0] for lemmas_tuple in lemmas_tuples
if lemmas_tuple[1] in lemmas_freqs
]
lemmas_list = ['<unk>'] + [w for w in wl if w in lemmas_list]
# Convert the list into hot vector dictionary
hv_dict = {}
for j, v in enumerate(lemmas_list):
hv_dict[v] = [int(i) for i in np.eye(len(lemmas_list), dtype=int)[j]]
with open(os.path.join(output_path, 'lemmas_dictionary.json'), 'w') as f:
json.dump(hv_dict, f)
def load_textgrid(file_name, tier_name='phone'):
'''Load textgrid & return times and labels'''
tg = tgt.read_textgrid(file_name)
tier = tg.get_tier_by_name(tier_name)
times = []
labels = []
for t in tier:
times.append([round(t.start_time, 4), round(t.end_time, 4)])
labels.append(t.text)
assert len(times) > 0, f'"times" is empty: len={len(times)}'
assert len(labels) > 0, f'"{tier_name}" is empty: len={len(labels)}'
return np.array(times, dtype='float32'), labels
def f1(predictions, grid_file, error=(0, 0)):
#Testing(grid_file,cre4,3200*0.256,pred,3200,for_pos)
vals = Testing(tgt.read_textgrid(grid_file),
4,
32000 * 0.256,
pickle.load(open(predictions, 'rb')),
32000,
for_pos=False)
false_pos, true_pos, merged_pos = find(
Testing(tgt.read_textgrid(grid_file), 4, 32000 * 0.256,
pickle.load(open(predictions, 'rb')), 32000, True), error)
if error != (0, 0):
vals = (delete_vals(vals[0], merged_pos), vals[1])
false_neg, true_neg = find(vals, (0, 0), for_pos=False)
#false_neg = true_neg = 1
print(false_pos, true_pos, false_neg, true_neg)
precision = true_pos / (true_pos + false_pos)
recall = true_pos / (true_pos + false_neg)
print(precision, recall)
score = 2 * (precision * recall / (precision + recall))
print(score)
return merged_pos
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 open_intervalframe_from_textgrid(filepath,
encoding='utf-8',
asobjects=False,
include_empty_intervals=False):
"""Import a textgrid and return a dict of IntervalFrames.
Each tier in the textgrid becomes an IntervalFrame (Pandas DataFrame)
The Intervals by default are tokenized into start_time, end_time and
text columns.
The points (for point tiers) are tokenized into time and mark columns.
Arguments:
filepath -- Path + filename of the TextGrid file to be imported.
Keyword Arguments:
asobjects -- If True, then values are intervalobjects (as defined in
package tgt, instead of tokenizing into start_time etc.
IntervalFrame has only one column with these objects.
include_empty_intervals -- If enabled, empty intervals between
annotations are also returned
encoding -- character encoding to read the textgrid file
"""
textgrid = tgt.read_textgrid(filepath, encoding, include_empty_intervals)
result = {}
for tier in textgrid.tiers:
if len(tier) > 0:
if isinstance(tier, tgt.IntervalTier):
frame = pd.DataFrame(tier.intervals, columns=['intervals'])
if asobjects == False:
frame['start_time'] = frame['intervals'].map(lambda x:\
x.start_time)
frame['end_time'] = frame['intervals'].map(lambda x:\
x.end_time)
frame['text'] = frame['intervals'].map(lambda x: x.text)
del frame['intervals']
elif isinstance(tier, tgt.PointTier):
frame = pd.DataFrame(tier.points, columns=['points'])
if asobjects == False:
frame['time'] = frame['points'].map(lambda x: x.time)
frame['mark'] = frame['points'].map(lambda x: x.text)
del frame['points']
result[tier.name] = frame
return result
def concatenate_textgrids(input_files, encoding):
"""Concatenate Tiers with matching names. TextGrids are concatenated
in the order they are specified. The number and the names of tiers
must be the same in each TextGrid."""
# Read all TextGrids into a list.
textgrids = [tgt.read_textgrid(path, encoding) for path in input_files]
# Check whether the TextGrids have the same number of tiers.
ntiers = [len(x) for x in textgrids]
assert all([ntiers[0] == x for x in ntiers[1:]]),\
'TextGrids have different numbers of tiers.'
# Check whether tiers in the TextGrids have the same names.
tier_names = [sorted(x.get_tier_names()) for x in textgrids]
assert all([tier_names[0] == x for x in tier_names[1:]]),\
'Names of tiers do not match.'
tot_duration = 0
tiers = {} # tier_name : tgt.Tier()
for textgrid in textgrids:
for tier in textgrid.tiers:
intervals = []
# If this is the first we see this tier, we just make a copy
# of it as it is.
if tier.name not in tiers.keys():
tiers[tier.name] = copy.deepcopy(tier)
# Otherwise we update the start and end times of intervals
# and append them to the first part.
else:
for interval in tier.intervals:
interval.left_bound += tot_duration
interval.right_bound += tot_duration
intervals.append(interval)
tiers[tier.name].add_intervals(intervals)
tot_duration += textgrid.end_time()
# Create a new TextGrid
textgrid_concatenated = tgt.TextGrid()
# Add tiers in the order they're found in the first TextGrid.
textgrid_concatenated.add_tiers([tiers[x] for x in textgrids[0].get_tier_names()])
return textgrid_concatenated
def open_intervalframe_from_textgrid(filepath, encoding='utf-8',
asobjects=False,
include_empty_intervals=False):
"""Import a textgrid and return a dict of IntervalFrames.
Each tier in the textgrid becomes an IntervalFrame (Pandas DataFrame)
The Intervals by default are tokenized into start_time, end_time and
text columns.
The points (for point tiers) are tokenized into time and mark columns.
Arguments:
filepath -- Path + filename of the TextGrid file to be imported.
Keyword Arguments:
asobjects -- If True, then values are intervalobjects (as defined in
package tgt, instead of tokenizing into start_time etc.
IntervalFrame has only one column with these objects.
include_empty_intervals -- If enabled, empty intervals between
annotations are also returned
encoding -- character encoding to read the textgrid file
"""
textgrid = tgt.read_textgrid(filepath, encoding, include_empty_intervals)
result = {}
for tier in textgrid.tiers:
if len(tier) > 0:
if isinstance(tier, tgt.IntervalTier):
frame = pd.DataFrame(tier.intervals, columns=['intervals'])
if asobjects == False:
frame['start_time'] = frame['intervals'].map(lambda x:\
x.start_time)
frame['end_time'] = frame['intervals'].map(lambda x:\
x.end_time)
frame['text'] = frame['intervals'].map(lambda x: x.text)
del frame['intervals']
elif isinstance(tier, tgt.PointTier):
frame = pd.DataFrame(tier.points, columns=['points'])
if asobjects == False:
frame['time'] = frame['points'].map(lambda x: x.time)
frame['mark'] = frame['points'].map(lambda x: x.text)
del frame['points']
result[tier.name] = frame
return result
def segment_textgrid_audio(textgrid_path, tier):
directory = os.listdir(textgrid_path)
try:
os.mkdir(textgrid_path + '/segments')
except:
print("Path already exists")
for f in directory:
if '.TextGrid' in f:
text_grid = tgt.read_textgrid(os.path.join(textgrid_path, f))
wav = os.path.join(textgrid_path, f.replace('TextGrid', 'wav'))
interval_num = 0
for interval in text_grid.tiers[tier].intervals:
interval_num +=1
if interval.text != '':
start_audio = interval.start_time * 1000
end_audio = interval.end_time * 1000
audio_segment = AudioSegment.from_file(wav)
audio_segment = audio_segment[start_audio:end_audio]
audio_segment.export(textgrid_path + '/segments/'+f.replace('.TextGrid', '_')+'_'+str(interval_num)+'.wav', format="wav")
def read_textgrid(filename, sample_rate=200):
import tgt
try:
tg = tgt.read_textgrid(filename) #, include_empty_intervals=True)
except:
print("reading " + filename + " failed")
return
tiers = []
labs = {}
for tier in tg.get_tier_names():
if (tg.get_tier_by_name(tier)).tier_type() != 'IntervalTier':
continue
tiers.append(tg.get_tier_by_name(tier))
lab = []
for a in tiers[-1].annotations:
try:
# this was for some past experiment
if a.text in ["p1", "p2", "p3", "p4", "p5", "p6", "p7"]:
lab[-1][-1] = lab[-1][-1] + "_" + a.text
else:
#lab.append([a.start_time*sample_rate,a.end_time*sample_rate,a.text.encode('utf-8')])
lab.append([
a.start_time * sample_rate, a.end_time * sample_rate,
a.text
])
except:
pass
#print tiers[-1].encode('latin-1')
labs[tier.lower()] = lab
try:
for i in range(len(labs['prosody'])):
if labs['prosody'][i][2][-2:] not in [
"p1", "p2", "p3", "p4", "p5", "p6", "p7"
]:
labs['prosody'][i][2] += "_p0"
except:
pass
return labs
def getAmatrix(TextGrid_Directory, CsvDirectory, N):
A = np.zeros((N, N))
# pdb.set_trace()
for root, dirs, filenames in os.walk(TextGrid_Directory):
# pdb.set_trace()
for f in filenames:
if f == '.DS_Store':
continue
tg = tgt.read_textgrid(TextGrid_Directory+f)
ipu_tier = tg.get_tier_by_name('Key-Child')
# pdb.set_trace()
prev = map(ipu_tier[0].text)
for i in range(1, len(ipu_tier)):
cur = map(ipu_tier[i].text)
if cur >= N:
break
A[prev, cur] += 1
prev = cur
return A
def getBmatrix_multi(num_label, TextGrid_Directory, TextGridname, CsvDirectory):
tg = tgt.read_textgrid(TextGrid_Directory+TextGridname)
ipu_tier = tg.get_tier_by_name('Key-Child')
filename = TextGridname.split('.')[0]
# dictionaries with key of filename, values of softmax probability predictions being that label
prob = {}
with open(CsvDirectory) as csvfile:
reader = csv.reader(csvfile, delimiter=' ', quotechar='|')
# prob_filenames = [row[0].split(',')[0] for row in reader]
for row in reader:
if row[0].split(',')[0] == 'total_filename':
continue
if num_label == 4:
prob[row[0].split(',')[0]] = [float(row[0].split(',')[-4]), float(row[0].split(',')[-3]), float(row[0].split(',')[-2]), float(row[0].split(',')[-1])]
else:
prob[row[0].split(',')[0]] = [float(row[0].split(',')[-5]), float(row[0].split(',')[-4]), float(row[0].split(',')[-3]), float(row[0].split(',')[-2]), float(row[0].split(',')[-1])]
y = []
prob_order = []
# append the probabilities of that segment being different classes in order
for seg in ipu_tier:
start_time = seg.start_time
end_time = seg.end_time
annotation = seg.text
segment_filename = filename + '-' + str(start_time) + '-' + str(end_time) + '-' + annotation
if num_label == 4:
if annotation == 'HIC':
continue
if not segment_filename in prob:
continue
if segment_filename in prob:
prob_order.append(prob[segment_filename])
y.append(map(annotation))
prob_order = np.array(prob_order)
res = prob_order.T
prediction_orignal = np.argmax(res, 0)
accuracy = sum(y == prediction_orignal) * 1.0/len(y)
FSCORE = f1_score(y, prediction_orignal, average='macro')
# print('original accuracy:',accuracy)
return res, accuracy, y, FSCORE
def get_tiernames_from_tgfile(read_file, print_encoding_info=False):
encoding_types = [
'utf-8', 'utf-8-sig', 'utf-16', 'utf-16-le', 'utf-16-be'
]
for encoding_type in encoding_types:
try:
tg = tgt.read_textgrid(read_file, encoding=encoding_type)
if type(tg) is tgt.core.TextGrid:
if print_encoding_info:
print(
f'\t\'{basename(read_file)}\' encoding in {encoding_type}.'
)
break
except:
# print(f'WARNING: failed with reading as {encoding_type}')
continue
try:
tier_names = [tier.name for tier in tg.tiers]
except:
print(f'ERROR: {read_file} not in {encoding_types}')
raise
return tg, tier_names
def __init__(self, path): # pp = pprint.PrettyPrinter(indent=4) self.generator = MarkovGenerator(n=1, max=3) self.path = path + '/textGrids/' #let's get the text from the textGrids, save the annotations in a dict, key=filename self.annotations = dict() for tgFile in os.listdir(self.path): if tgFile[-9:] == '.TextGrid': #print tgFile tg = tgt.read_textgrid(self.path + tgFile) file_annotations = [i for t in tg.tiers for i in t] for i in range(len(file_annotations)): a1 = file_annotations[i] filename = tgFile[:-9] self.annotations[a1.text] = (filename, a1) if i == len(file_annotations)-1: continue else: a2 = file_annotations[i+1] self.feedMarkov(a1,a2)
def main():
args = parser()
tgfiles = args.textgrid.split(',')
identifiers = args.identifier.split(',')
assert len(tgfiles) == len(identifiers), "number of textgrids must match number of identifiers"
pairs = zip(tgfiles, identifiers)
rows = []
for tgfile, identifier in pairs:
tg = tgt.read_textgrid(tgfile)
tier = tg.get_tier_by_name(args.tier)
matches = tier.get_annotations_with_text(args.regex, regex=True)
for m in matches:
rows.append((str(m.start_time), str(m.end_time), str(m.duration()), m.text, identifier))
with open(args.output_path, 'w') as out:
out.write("start\tend\tduration\tlabel\tidentifier\n")
for row in rows:
out.write('\t'.join(row) + '\n')
def calculate_durations(textgrid, phone2idx):
tokens = []
durs = []
frames_per_second = args.sr / args.hop_length
tg = tgt.read_textgrid(textgrid, include_empty_intervals=True)
data_tier = tg.get_tier_by_name("phones")
# Get total frames
total_frames = ceil((data_tier.end_time - data_tier.start_time) * frames_per_second)
# Find start and end frames of each token
se_in_frames = np.array([(frames_per_second * d.start_time, frames_per_second * d.end_time) for d in data_tier])
se_in_frames = np.round(se_in_frames)
durs = (se_in_frames[:, 1] - se_in_frames[:, 0]).astype(int)
blank_set = ('sil', 'sp', 'spn', '', '<unk>')
blank_token = " "
# merge repeated blank tokens
tokens, durations = [], []
for i in range(len(data_tier)):
x = data_tier[i].text
if x == 'spn':
return None, None, None
x = blank_token if x in blank_set else x
if len(tokens) and tokens[-1] == blank_token and x == blank_token:
durations[-1] += durs[i]
else:
tokens.append(x)
durations.append(durs[i])
tokens_enc = [phone2idx[token] for token in tokens]
tokens_enc, durations = torch.LongTensor(tokens_enc), torch.LongTensor(durations)
# Add rounding error to final token
durations[-1] += total_frames - durations.sum()
return tokens, tokens_enc, durations
def get_textgrid_sa(mfa_file):
read_textgrid = tgt.read_textgrid(mfa_file)
[words, start_time, end_time] = read_word_alignment(read_textgrid)
assert len(words) == len(start_time) == len(end_time)
stack = []
for i in range(len(words)):
if words[i] == "":
continue
if stack:
if start_time[i] > stack[-1][-1]:
stack.append([start_time[i], end_time[i]])
#print("non contiguos word")
else:
stack[-1][-1] = end_time[i]
else:
stack.append([start_time[i], end_time[i]])
return stack
def main():
args = parser()
tgfiles = args.textgrid.split(',')
identifiers = args.identifier.split(',')
assert len(tgfiles) == len(
identifiers), "number of textgrids must match number of identifiers"
pairs = zip(tgfiles, identifiers)
rows = []
for tgfile, identifier in pairs:
tg = tgt.read_textgrid(tgfile)
tier = tg.get_tier_by_name(args.tier)
matches = tier.get_annotations_with_text(args.regex, regex=True)
for m in matches:
rows.append((str(m.start_time), str(m.end_time), str(m.duration()),
m.text, identifier))
with open(args.output_path, 'w') as out:
out.write("start\tend\tduration\tlabel\tidentifier\n")
for row in rows:
out.write('\t'.join(row) + '\n')
def textgrid2rttm(textgrid):
'''
Take in input the path to a text grid,
and output a dictionary of lists *{spkr: [ (onset, duration) ]}*
that can easily be written in rttm format.
'''
# init output
rttm_out = dict()
# open textgrid
#tg = tgio.openTextgrid(textgrid)
tg = tgt.read_textgrid(textgrid)
# loop over all speakers in this text grid
#for spkr in tg.tierNameList:
for spkr in tg.get_tier_names():
spkr_timestamps = []
# loop over all annotations for this speaker
#for interval in tg.tierDict[spkr].entryList:
for _interval in tg.get_tiers_by_name(spkr):
for interval in _interval:
bg, ed, label = interval.start_time,\
interval.end_time,\
interval.text
#if label == "x":
# continue
#elif label == "1" or label == "2":
# spkr_timestamps.append((bg, ed-bg))
spkr_timestamps.append((bg, ed-bg))
# add list of onsets, durations for each speakers
rttm_out[spkr] = spkr_timestamps
return rttm_out
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 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')
for j in intervals:
if i == j.text:
intervals.remove(j)
break
# intervals = [interval for interval in intervals if interval.text not in string.ascii_uppercase[:4]]
return intervals
def standardize(word):
word = re.sub(r'\[.+?\]\s*', '', word)
return word
#handle the textgrid file
files = [file for file in os.listdir(folder) if file.endswith(".TextGrid")]
for f in files:
tg = tgt.read_textgrid(f)
tier = tg.get_tier_by_name('words')
#get intervals
intervals = [i for i in tier.intervals if i.text != 'sp' and i.text != 'sil']
intervals = remove_redundant(intervals)
#match the file
index_json = match_file(f[:len(f)-9])
#update json
json_type = json.loads(json_list[index_json])
count = 0
print f
for i in json_type["imgs"]:
for j in i["texts"]:
words = j["content"].split(" ")
def setUp(self):
self.f = open_streamframe_from_xiofile(
"data/fseeksmaller.xio.gz",
"lab-labtop/irioKinect 2",
window_size=5,
with_fields=[],
without_fields=[],
discard_duplicates=True,
start_time=0,
end_time=13,
relative=True,
timestamp_offset=10,
)
self.f2 = open_streamframe_from_xiofile(
"data/fseeksmaller.xio.gz",
"lab-labtop/irioKinect",
window_size=5,
with_fields=[],
without_fields=[],
discard_duplicates=True,
start_time=0,
end_time=13,
relative=True,
timestamp_offset=10,
)
# self.outtake_from_stream = self.f.ix[1341393414826]['framenumber']
self.fraw = open_streamframe_from_xiofile(
"data/fseeksmaller.xio.gz",
"lab-labtop/irioKinect 2",
window_size=5,
with_fields=[],
without_fields=[],
discard_duplicates=True,
start_time=0,
end_time=13,
relative=True,
timestamp_offset="raw",
)
self.ff = open_streamframe_from_xiofile(
"data/fseeksmaller.xio.gz",
"lab-labtop/irioKinect 2",
window_size=5,
with_fields=[],
without_fields=[],
discard_duplicates=True,
start_time=0,
end_time=13,
relative=True,
)
save_streamframe_to_xiofile({"lab-labtop/irioKinect 2": self.f}, "data/sf_to_xio.xio.gz")
save_streamframe_to_xiofile(
{"lab-labtop/irioKinect 2": self.f, "lab-labtop/irioKinect": self.f2}, "data/sf_to_xio2.xio.gz"
)
self.rsn = open_streamframe_from_xiofile("data/fseeksmaller.xio.gz", "wrong/sensor/name")
self.outtake_from_xio = XIOFile("data/sf_to_xio.xio.gz", indexing=True)
self.outtake_from_xio_2 = XIOFile("data/sf_to_xio2.xio.gz", indexing=True)
self.q = [
ex for ex in quantize(self.outtake_from_xio.xio_quicklinegen(0, 13, True, True), "lab-labtop/irioKinect 2")
][0]["soundAngle"]
self.ivf = open_intervalframe_from_textgrid(
"data/r1_12_15with" "Point.TextGrid", encoding="utf-8", asobjects=False, include_empty_intervals=False
)["P"]
self.cv = convert_pointtier_to_streamframe(self.ivf)
self.pf = convert_streamframe_to_pointtier(self.f)
self.outtake_from_pf = convert_streamframe_to_pointtier(self.f)["soundAngle"].ix[0]["time"]
self.if_from_tg = open_intervalframe_from_textgrid(
"data/r1-20120704-" "cam1-head-zm.TextGrid",
encoding="utf-8",
asobjects=False,
include_empty_intervals=False,
)
self.if_from_tg_tier = self.if_from_tg.values()[0]
save_intervalframe_to_textgrid(self.if_from_tg, "data/testif.TextGrid", encoding="utf-8")
self.tg = tgt.read_textgrid("data/testif.TextGrid", encoding="utf-8", include_empty_intervals=False)
self.ic1 = open_intervalframe_from_increco("data/test.inc_reco")
self.ic2 = open_intervalframe_from_increco("data/test.inc_reco", lastonly=True)
def main(wav_path, outfile_path, speech_path=None, speech_tier_name=None):
if speech_path is not None and speech_path is not None:
tg = tgt.read_textgrid(speech_path)
speech = tg.get_tier_by_name(speech_tier_name)
else:
speech = None
# Read the respiratory signal, detrend it, identify cycles and
# holds.
resp = rip.RIP.from_wav(wav_path, speech=speech)
resp.remove_baseline()
resp.find_cycles(include_holds=True)
resp.find_holds()
resp.estimate_range()
resp.estimate_rel(30)
resp.save_annotations('breath.TextGrid')
fname = os.path.splitext(os.path.basename(wav_path))[0]
# For each inhalation and exhalation, extract the respiratory
# cycles.
features = []
for i, seg in enumerate(resp.segments):
start = seg.start_time
end = seg.end_time
label = seg.text
features_seg = {
'file': fname,
'start': start,
'end': end,
'segment': label
}
# Odd-numbered rows correspond to inhalations and even-numbered
# rows correspond to exhalations.
if label == 'out':
cycle_start = resp.segments[i - 1].start_time
features_seg['duty_cycle'] = (end - start) / (end - cycle_start)
else:
cycle_end = resp.segments[i + 1].end_time
features_seg['duty_cycle'] = (end - start) / (cycle_end - start)
features_seg['duration'] = end - start
features_seg['slope'] = resp.extract_slope(start, end)
features_seg['amplitude'] = resp.extract_amplitude(start, end)
features_seg['vol_start'] = resp.extract_level(start)
features_seg['vol_end'] = resp.extract_level(end)
holds = resp.holds.get_annotations_between_timepoints(
start, end, left_overlap=True, right_overlap=True)
features_seg['nholds'] = len(holds)
if len(holds):
holds_dur = sum(h.end_time - h.start_time for h in holds)
holds_dur -= max(0, start - holds[0].start_time)
holds_dur -= max(0, holds[-1].end_time - end)
else:
holds_dur = 0
if speech is not None:
# Extract: time lag, interval before and after: duration,
# start, end level, slope
pass
features.append(features_seg)
with open(outfile_path, 'w') as fout:
csv_out = csv.DictWriter(fout, fieldnames=features[0].keys())
csv_out.writeheader()
csv_out.writerows(features)
def duration(path, C_list, V_list, cid):
#file_list = glob.glob(path + r"\*\sent\*.TextGrid") # glob匹配所有的符合条件的文件，并将以list的形式返回
file_list = glob.glob(path + r"\sent\*.TextGrid") # glob匹配所有的符合条件的文件，并将以list的形式返回
#print(file_list)
AlldeltS = [] # 依次计算 每一个 textgrid 的 结果值 把结果存在 总列表中
all_vs = []
all_rpvis = [] # 可能有点问题 每次累加进去 一个值 但是 不清空
all_npvis = []
all_ms = []
for file in file_list:
TextGrid = tgt.read_textgrid(file, include_empty_intervals=True) # 依次读取TextGrid文件
if cid == 'jp':
tier = TextGrid.get_tier_by_name(TextGrid.get_tier_names()[0])
#print(tier)
if cid == 'cn':
tier = TextGrid.get_tier_by_name(TextGrid.get_tier_names()[0])
#print(tier)
elif cid == 'ru':
tier = TextGrid.get_tier_by_name(TextGrid.get_tier_names()[1]) # 根据 tier的 name/位置 读取 intervals
#print(tier)
# tier = TextGrid.get_tier_by_name('SY')
tier_name = TextGrid.get_tier_names() # 获取全部的tier 名字
start = tier.start_time
end = tier.end_time
start_syl = tier.start_time
end_syl = tier.end_time
tier2insert = tgt.IntervalTier(start, end, name='CV') # 获取起始点和 终点 插入一条 CV的 intervals
TextGrid.insert_tier(tier2insert, 3)
CV = TextGrid.get_tier_by_name('CV')
annotation = tier.intervals # 插入一个 intervals
#syllable = tier_syll.intervals
num = []
S_duration = [] # syllable_duration
duration_all_S = 0 # 全部时长和 （用于计算 %V 和 其他相关参数）
for i in range(len(annotation)): # 循环 替换 和 计算 时长
old_name = annotation[i].text
old_start_time = annotation[i].start_time
old_end_time = annotation[i].end_time
duration = old_end_time - old_start_time
#if old_name in C_list: # 判断 属于 C / V
if old_name != 'sil':
new_name = 'S'
# elif old_name in V_list: # 判断 属于 C / V
# new_name = 'S'
else:
new_name = 'none'
# print(old_name, new_name)
Interval = tgt.Interval(old_start_time, old_end_time, text=new_name) # interval格式- 依次填写
# print(old_name, new_name, 'duration=', duration)
if new_name == 'S':
S_duration.append(duration) # 加入 duration 的list
duration_all_S = duration_all_S + duration
CV.add_interval(Interval) # 将 intervals 的标注 >> 到 textgrid
#print(file, S_duration)
mean_syl = duration_all_S/len(S_duration)
#print(mean_syl)
vacS = duration_all_S / len(S_duration)
# print(num)
# if num > 0:
# mean_syl = a / (len(C_duration) + len(V_duration)) # 计算一个 mean_syllable duration 用于 语速
# print(mean_syl)
# mean_syl = a/(len(C_duration)+len(V_duration))
# print(mean_syl)
# vacroC = round(deltaC(C_duration) / mean_syl * 100, 4)
# vacroV = round(deltaC(V_duration) / mean_syl * 100, 4)
vacroS = round(deltaS(S_duration) / vacS * 100, 4)
# print(file, ',',
#
# deltaS(S_duration), ',',
#
# vacroS, ',',
#
# rPVI_s(S_duration), ',',
#
# nPVI_S(S_duration), ',',
# )
#print(nPVI_S(S_duration))
AlldeltS.append(deltaS(S_duration))
all_vs.append(vacroS)
all_rpvis.append(rPVI_s(S_duration))
all_npvis.append(nPVI_S(S_duration))
all_ms.append(mean_syl)
deltS = round(np.mean(AlldeltS), 9)
vs = round(np.mean(all_vs), 9)
rpvis = round(np.mean(all_rpvis), 9)
npvis = round(np.mean(all_npvis), 9)
ms = round(np.mean(all_ms), 9)
#
print(path, ',',
ms, ',',
deltS, ',',
vs, ',',
rpvis, ',',
npvis, ',',
)
