def get_cross_talk(database_path, dataset, json_path):
database_path = database_path / 'CHiME5'
dataset_transciption_path = database_path / 'transcriptions' / dataset
json_dict = dict()
for session_path in dataset_transciption_path.glob('*.json'):
trans = load_json(session_path)
session_id = trans[0]['session_id']
json_dict[session_id] = dict()
total = len(trans)
speaker_ids = [
key for key in trans[0]['start_time'].keys() if 'P' in key
]
out_dict = {
speaker:
{speaker: dict(start=[], end=[])
for speaker in speaker_ids}
for speaker in speaker_ids
}
from concurrent.futures import ThreadPoolExecutor
with ThreadPoolExecutor(os.cpu_count()) as ex:
for speaker, example_dict in tqdm.tqdm(
ex.map(get_dict_speaker, trans),
total=total,
desc=dataset + '_' + session_id):
if example_dict is not None:
out_dict[speaker] = combine_dicts(example_dict,
out_dict[speaker])
out_dict['cross_talk'] = get_cross_talk_per_mic(out_dict)
dump_json(out_dict, str(json_path / session_id) + '.json')
def load_transciption_json(path, chime6):
try:
return transform_transciption_list(load_json(path), chime6)
except Exception as e:
raise RuntimeError(
'See above exception msg.\n'
f'The problematic json file is {path}.'
) from e
def get_active_speaker(start_sample,
end_sample,
session_id,
mic_id,
json_path=None,
speaker_json=None,
sample_step=1,
dtype=bool):
if json_path is not None:
speaker_json = load_json(str(json_path / session_id) + '.json')
elif speaker_json is None:
raise ValueError('Either json_path or speaker_json have to be defined')
out_dict = dict()
for key, value in speaker_json['cross_talk'].items():
cross_talk = to_numpy(value,
start_sample,
end_sample,
sample_step=sample_step,
dtype=dtype)
activity = to_numpy(speaker_json[key][mic_id], start_sample,
end_sample, sample_step, dtype)
out_dict[key] = dict(cross_talk=cross_talk, activity=activity)
return out_dict
def load_transciption_json(path):
return transform_transciption_list(load_json(path))