def read_eaf(ie, tier, silence_tier, silence_marker, json_data, output_text_dir, output_audio_dir):
input_eaf = Eaf(ie)
# Check if the tiers we have been given exist
tier_names = input_eaf.get_tier_names()
if tier not in tier_names:
print('missing tier: ' + tier, file=sys.stderr)
return False
if silence_tier not in tier_names:
print('missing silence tier: ' + silence_tier, file=sys.stderr)
# get the input_scripts audio file
inDir, name = os.path.split(ie)
basename, ext = os.path.splitext(name)
ia = os.path.join(inDir, basename + ".wav")
input_audio = AudioSegment.from_wav(ia)
# We can pass in an arg for a ref tier that has silence labels
check_silence_ref_tier = False
if silence_tier in tier_names:
silence_tier_info = input_eaf.get_parameters_for_tier(silence_tier)
if silence_tier_info.get("PARENT_REF") == tier:
check_silence_ref_tier = True
# Get annotation values, start and end times, and speaker id
annotations = sorted(input_eaf.get_annotation_data_for_tier(tier))
params = input_eaf.get_parameters_for_tier(tier)
if 'PARTICIPANT' in params:
speaker_id = params['PARTICIPANT']
i = 0
for ann in annotations:
skip = False
start = ann[0]
end = ann[1]
# output_scripts new values, not the original clip start end times
clip_start = 0
clip_end = ann[1] - ann[0]
annotation = ann[2]
# Check for annotations labelled with a particular symbol on the main tier
if annotation == silence_marker:
skip = True
# Check for existence of an annotation in ref tier to silence
# Annotation value doesn't matter
if check_silence_ref_tier and len(input_eaf.get_ref_annotation_at_time(silence_tier, start)):
skip = True
if skip is True:
# print('skipping annotation: ' + annotation, start, end)
print("skipping" + str(i))
else:
print("processing" + str(i))
# print('processing annotation: ' + annotation, start, end)
# build the output_scripts audio/text filename
fname = basename + "_" + str(i)
obj = {
'audioFileName': os.path.join(".", fname + ".wav"),
'transcript': annotation,
'startMs': clip_start,
'stopMs': clip_end
}
if 'PARTICIPANT' in params:
obj["speakerId"] = speaker_id
json_data.append(obj)
split_audio_by_start_end(input_audio, start, end, fname, ".wav", output_audio_dir)
write_text(annotation, fname, ".txt", output_text_dir)
i += 1
def read_eaf(ie):
if verbose:
print("input file is", ie)
input_eaf = Eaf(ie)
# Check if the tiers we have been given exist
tier_names = list(input_eaf.get_tier_names())
if verbose:
print("tier_names", tier_names, file=sys.stderr)
# Are we working by slice_tier name or order?
if slice_tier != "default":
if verbose:
print("using slice_tier by name:", slice_tier, file=sys.stderr)
else:
# Sanity check that the slice_tier num is not greater than the num of tiers
if tier_order > len(tier_names):
print("Error: tier number is greater than the number of tiers",
file=sys.stderr)
return False
if verbose:
print("using slice_tier by number:",
tier_names[tier_order - 1],
file=sys.stderr)
if slice_tier not in tier_names:
print('Error: missing slice_tier ' + slice_tier, file=sys.stderr)
return False
if silence_tier not in tier_names:
if verbose:
print('silence tier not found: ' + silence_tier, file=sys.stderr)
# get the input audio file
inDir, name = os.path.split(ie)
basename, ext = os.path.splitext(name)
# we can write out mp3 or whatever, still require wav input
ia = os.path.join(inDir, basename + ".wav")
input_audio = AudioSegment.from_wav(ia)
# We can pass in an arg for a ref tier that has silence labels
check_silence_ref_tier = False
if silence_tier in tier_names:
silence_tier_info = input_eaf.get_parameters_for_tier(silence_tier)
if silence_tier_info.get("PARENT_REF") == tier:
check_silence_ref_tier = True
# Get annotation values, start and end times, and speaker id
if text_tier not in tier_names:
print('Error: missing text tier')
return False
annotations = sorted(input_eaf.get_annotation_data_for_tier(text_tier))
params = input_eaf.get_parameters_for_tier(text_tier)
if 'PARTICIPANT' in params:
speaker_id = params['PARTICIPANT']
annotations_data = []
i = 0
for ann in annotations:
skip = False
ref_annotation = []
start = ann[0]
end = ann[1]
# output new values, not the original clip start end times
clip_start = 0
clip_end = ann[1] - ann[0]
annotation = ann[2]
# Check for annotations labelled with a particular symbol on the main tier
if annotation == silence_marker:
skip = True
# Check for existence of an annotation in ref tier to silence
# Annotation value doesn't matter
if check_silence_ref_tier:
ref_annotation = input_eaf.get_ref_annotation_at_time(
silence_tier, start)
if len(ref_annotation) is True:
skip = True
if skip is True:
print('skipping annotation: ' + annotation, start, end)
else:
print('processing annotation: ' + annotation, start, end)
# build the output audio/text filename
fname = basename + "_" + str(i)
if name_with_annotation:
fname = slugify(annotation)
if prefix != '':
fname = prefix + '_' + fname
obj = {
'audioFileName': os.path.join(".", fname + ".wav"),
'transcript': annotation,
'startMs': clip_start,
'stopMs': clip_end
}
if 'PARTICIPANT' in params:
obj["speakerId"] = speaker_id
annotations_data.append(obj)
split_audio_by_start_end(input_audio, start, end, fname)
write_text(annotation, fname)
i += 1
# output the json data for the next step in kaldi pipeline
write_json(annotations_data)
if verbose:
print(annotations_data)
