def convert(file_id_list,
in_lab_dir,
in_feats_dir,
fs,
out_lab_dir,
b_prevent_zeros=False):
'''
b_prevent_zeros: True if you want to ensure that all the phonemes have one frame at least.
(not recommended, only useful when there are too many utterances crashed)
'''
# Conversion:
lu.mkdir(out_lab_dir)
v_filenames = lu.read_text_file2(file_id_list,
dtype='string',
comments='#')
crashlist_file = lu.ins_pid('crash_file_list.scp')
for filename in v_filenames:
# Display:
print('\nConverting lab file: ' + filename +
'................................')
# Current i/o files:
in_lab_file = os.path.join(in_lab_dir, filename + '.lab')
out_lab_file = os.path.join(out_lab_dir, filename + '.lab')
in_shift_file = os.path.join(in_feats_dir, filename + '.shift')
# Debug:
'''
v_shift = lu.read_binfile(in_shift_file, dim=1)
v_n_frms = mp.get_num_of_frms_per_state(v_shift, in_lab_file, fs, b_prevent_zeros=b_prevent_zeros)
la.convert_label_state_align_to_var_frame_rate(in_lab_file, v_n_frms, out_lab_file)
#'''
v_n_frms = 0
try:
v_shift = lu.read_binfile(in_shift_file, dim=1)
v_n_frms = mp.get_num_of_frms_per_state(
v_shift,
in_lab_file,
fs,
b_prevent_zeros=b_prevent_zeros,
n_states_x_phone=1)
la.convert_label_state_align_to_var_frame_rate(
in_lab_file, v_n_frms, out_lab_file)
except (KeyboardInterrupt, SystemExit):
raise
except:
print("crashlist")
with open(crashlist_file, "a") as crashlistlog:
crashlistlog.write(filename + '\n')
print('Done!')
def convert(file_id_list, in_lab_dir, in_feats_dir, fs, out_lab_dir, b_prevent_zeros=False):
'''
b_prevent_zeros: True if you want to ensure that all the phonemes have one frame at least.
(not recommended, only useful when there are too many utterances crashed)
'''
# Conversion:
lu.mkdir(out_lab_dir)
v_filenames = lu.read_text_file2(file_id_list, dtype='string', comments='#')
crashlist_file = lu.ins_pid('crash_file_list.scp')
for filename in v_filenames:
# Display:
print('\nConverting lab file: ' + filename + '................................')
# Current i/o files:
in_lab_file = os.path.join(in_lab_dir , filename + '.lab')
out_lab_file = os.path.join(out_lab_dir , filename + '.lab')
in_shift_file = os.path.join(in_feats_dir, filename + '.shift')
# Debug:
'''
v_shift = lu.read_binfile(in_shift_file, dim=1)
v_n_frms = mp.get_num_of_frms_per_state(v_shift, in_lab_file, fs, b_prevent_zeros=b_prevent_zeros)
la.convert_label_state_align_to_var_frame_rate(in_lab_file, v_n_frms, out_lab_file)
#'''
try:
v_shift = lu.read_binfile(in_shift_file, dim=1)
v_n_frms = mp.get_num_of_frms_per_state(v_shift, in_lab_file, fs, b_prevent_zeros=b_prevent_zeros)
la.convert_label_state_align_to_var_frame_rate(in_lab_file, v_n_frms, out_lab_file)
except (KeyboardInterrupt, SystemExit):
raise
except:
with open(crashlist_file, "a") as crashlistlog:
crashlistlog.write(filename + '\n')
print('Done!')
if __name__ == '__main__':
# CONSTANTS: So far, the vocoder has been tested only with the following constants:===
fs = 48000
# INPUT:==============================================================================
files_scp = '../data/file_id.scp' # List of file names (tokens). Format used by Merlin.
in_lab_st_dir = '../data/labs' # Original state aligned label files directory (in the format used by Merlin).
in_shift_dir = '../data/params' # Directory containing .shift files (You need to run feature extraction before running this script.)
out_lab_st_dir = '../data/labs_var_rate' # Directory that will contain the converted "variable frame rate" state aligned label files.
b_prevent_zeros = False # True if you want to ensure that all the phonemes have one frame at least. (not recommended, only usful when there are too many utterances crashed)
# PROCESSING:=========================================================================
lu.mkdir(out_lab_st_dir)
v_fileTokns = lu.read_text_file2(files_scp, dtype='string', comments='#')
n_files = len(v_fileTokns)
crashlist_file = lu.ins_pid('crash_file_list.scp')
for ftkn in v_fileTokns:
# Display:
print('\nAnalysing file: ' + ftkn + '................................')
# Input files:
in_lab_st_file = in_lab_st_dir + '/' + ftkn + '.lab'
out_lab_st_file = out_lab_st_dir + '/' + ftkn + '.lab'
in_shift_file = in_shift_dir + '/' + ftkn + '.shift'
try:
v_shift = lu.read_binfile(in_shift_file, dim=1)
if __name__ == '__main__':
# CONSTANTS: So far, the vocoder has been tested only with the following constants:===
fs = 48000
# INPUT:==============================================================================
files_scp = '../data_48k/file_id.scp' # List of file names (tokens). Format used by Merlin.
in_lab_st_dir = '../data_48k/labs' # Original state aligned label files directory (in the format used by Merlin).
in_shift_dir = '../data_48k/params' # Directory containing .shift files (You need to run feature extraction before running this script.)
out_lab_st_dir = '../data_48k/labs_var_rate' # Directory that will contain the converted "variable frame rate" state aligned label files.
b_prevent_zeros = False # True if you want to ensure that all the phonemes have one frame at least. (not recommended, only usful when there are too many utterances crashed)
# PROCESSING:=========================================================================
lu.mkdir(out_lab_st_dir)
v_fileTokns = lu.read_text_file2(files_scp, dtype='string', comments='#')
n_files = len(v_fileTokns)
crashlist_file = lu.ins_pid('crash_file_list.scp')
for ftkn in v_fileTokns:
# Display:
print('\nAnalysing file: ' + ftkn + '................................')
# Input files:
in_lab_st_file = in_lab_st_dir + '/' + ftkn + '.lab'
out_lab_st_file = out_lab_st_dir + '/' + ftkn + '.lab'
in_shift_file = in_shift_dir + '/' + ftkn + '.shift'
try:
v_shift = lu.read_binfile(in_shift_file, dim=1)
lu.write_binfile(v_lf0, out_feats_dir + '/' + file_name_token + '.lf0')
# Saving auxiliary feature shift (hop length). It is useful for posterior modifications of labels in Merlin.
lu.write_binfile(v_shift, out_feats_dir + '/' + file_name_token + '.shift')
return
if __name__ == '__main__':
# CONSTANTS: So far, the vocoder has been tested only with the following constants:===
fft_len = 4096
fs = 48000
# INPUT:==============================================================================
files_scp = '../data/file_id.scp' # List of file names (tokens). Format used by Merlin.
in_wav_dir = '../data/wavs_nat' # Directory with the wavfiles to extract the features from.
out_feats_dir = '../data/params' # Output directory that will contain the extracted features.
mvf = 4500 # Maximum voiced frequency (Hz)
# FILES SETUP:========================================================================
lu.mkdir(out_feats_dir)
l_file_tokns = lu.read_text_file2(files_scp, dtype='string',
comments='#').tolist()
# MULTIPROCESSING EXTRACTION:==========================================================
lu.run_multithreaded(feat_extraction, in_wav_dir, l_file_tokns,
out_feats_dir, fft_len, mvf)
print('Done!')
# File setup:
wav_file = os.path.join(in_wav_dir, file_name_token + '.wav')
mp.analysis_compressed(wav_file, out_dir=out_feats_dir)
return
if __name__ == '__main__':
# INPUT:==============================================================================
files_scp = '../data_48k/file_id.scp' # List of file names (tokens). Format used by Merlin.
in_wav_dir = '../data_48k/wavs_nat' # Directory with the wavfiles to extract the features from.
out_feats_dir = '../data_48k/params' # Output directory that will contain the extracted features.
# FILES SETUP:========================================================================
lu.mkdir(out_feats_dir)
l_file_tokns = lu.read_text_file2(files_scp, dtype='string', comments='#').tolist()
# MULTIPROCESSING EXTRACTION:==========================================================
lu.run_multithreaded(feat_extraction, in_wav_dir, l_file_tokns, out_feats_dir)
# For debug (Don't remove):
#for file_name_token in l_file_tokns:
# feat_extraction(in_wav_dir, file_name_token, out_feats_dir)
print('Done!')
print(
"\nCopying files from base experiment to current experiment location.............."
)
# Copy files and directories from base to current experiment:
copytree(base_exper_path, l_files_and_dirs_to_copy, exper_path)
os.rename(os.path.join(exper_path, 'conf/config.conf'),
os.path.join(exper_path, 'conf/config_base.conf'))
# Save backup of this file and used magphase code:
shutil.copytree(os.path.dirname(mp.__file__),
os.path.join(exper_path, 'backup_magphase_code'))
shutil.copy2(__file__, os.path.join(exper_path, 'conf'))
# Read file list:
l_file_tokns = lu.read_text_file2(os.path.join(exper_path, file_id_list),
dtype='string',
comments='#').tolist()
if b_feat_extr:
# Extract features:
acoustic_feats_path = os.path.join(exper_path, acoustic_feats_dir)
lu.mkdir(acoustic_feats_path)
if b_feat_ext_multiproc:
lu.run_multithreaded(feat_extraction, in_wav_dir, l_file_tokns,
acoustic_feats_path, d_mp_opts)
else:
for file_name_token in l_file_tokns:
feat_extraction(in_wav_dir, file_name_token,
acoustic_feats_path, d_mp_opts)
if __name__ == '__main__':
# Parsing input arg:
config_file = sys.argv[1]
# Constants:
b_prevent_zeros = False # True if you want to ensure that all the phonemes have one frame at least.
# (not recommended, only usful when there are too many utterances crashed)
# Parsing config file:
file_id_list, in_lab_dir, in_feats_dir, fs, out_lab_dir = parse_config_file(config_file)
# Conversion:
lu.mkdir(out_lab_dir)
v_filenames = lu.read_text_file2(file_id_list, dtype='string', comments='#')
n_files = len(v_filenames)
crashlist_file = lu.ins_pid('crash_file_list.scp')
for filename in v_filenames:
# Display:
print('\nConverting lab file: ' + filename + '................................')
# Current i/o files:
in_lab_file = path.join(in_lab_dir , filename + '.lab')
out_lab_file = path.join(out_lab_dir , filename + '.lab')
# Debug:
#print('out_lab_file:-----------------------------------')
