def feat_extraction_magphase(in_wav_dir, file_id_list, cfg, logger, b_multiproc=False):
sys.path.append(cfg.magphase_bindir)
import libutils as lu
import magphase as mp
def feat_extraction_magphase_one_file(in_wav_dir, file_name_token, acous_feats_dir, cfg, logger):
# Logging:
logger.info('Analysing waveform: %s.wav' % (file_name_token))
# File setup:
wav_file = os.path.join(in_wav_dir, file_name_token + '.wav')
# Feat extraction:
mp.analysis_for_acoustic_modelling(wav_file, out_dir=acous_feats_dir, mag_dim=cfg.mag_dim,
phase_dim=cfg.real_dim, b_const_rate=cfg.magphase_const_rate)
return
if b_multiproc:
lu.run_multithreaded(feat_extraction_magphase_one_file, in_wav_dir, file_id_list, cfg.acous_feats_dir, cfg, logger)
else:
for file_name_token in file_id_list:
feat_extraction_magphase_one_file(in_wav_dir, file_name_token, cfg.acous_feats_dir, cfg, logger)
return
def feat_extraction_magphase(in_wav_dir,
file_id_list,
cfg,
logger,
b_multiproc=False):
sys.path.append(cfg.magphase_bindir)
import libutils as lu
import magphase as mp
def feat_extraction_magphase_one_file(in_wav_dir, file_name_token,
acous_feats_dir, cfg, logger):
# Logging:
logger.info('Analysing waveform: %s.wav' % (file_name_token))
# File setup:
wav_file = os.path.join(in_wav_dir, file_name_token + '.wav')
# Feat extraction:
mp.analysis_for_acoustic_modelling(
wav_file,
out_dir=acous_feats_dir,
mag_dim=cfg.mag_dim,
phase_dim=cfg.real_dim,
b_const_rate=cfg.magphase_const_rate)
return
if b_multiproc:
lu.run_multithreaded(feat_extraction_magphase_one_file, in_wav_dir,
file_id_list, cfg.acous_feats_dir, cfg, logger)
else:
for file_name_token in file_id_list:
feat_extraction_magphase_one_file(in_wav_dir, file_name_token,
cfg.acous_feats_dir, cfg, logger)
return
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!')
# Read file list:
file_id_list = pars_acous_train['Paths']['file_id_list']
l_file_tokns = lu.read_text_file2(file_id_list,
dtype='string',
comments='#').tolist()
acoustic_feats_path = pars_acous_train['Paths']['in_acous_feats_dir']
# Acoustic Feature Extraction:-------------------------------------------------------------
if b_feat_extr:
# Extract features:
lu.mkdir(acoustic_feats_path)
if b_feat_ext_multiproc:
lu.run_multithreaded(
feat_extraction,
join(exper_path, 'acoustic_model', 'data', 'wav'),
l_file_tokns, acoustic_feats_path, d_mp_opts)
else:
for file_name_token in l_file_tokns:
feat_extraction(
join(exper_path, 'acoustic_model', 'data', 'wav'),
file_name_token, acoustic_feats_path, d_mp_opts)
# Labels Conversion to Variable Frame Rate:------------------------------------------------
if b_conv_labs_rate and not d_mp_opts[
'b_const_rate']: # NOTE: The script ./script/label_st_align_to_var_rate.py can be also called from comand line directly.
label_state_align = join(exper_path, 'acoustic_model', 'data',
'label_state_align')
label_state_align_var_rate = pars_acous_train['Labels']['label_align']
fs = int(pars_acous_train['Waveform']['samplerate'])
ltvr.convert(file_id_list, label_state_align, acoustic_feats_path, fs,
# 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!')
return
def get_wav_filelist(wav_dir):
wav_files = []
for file in os.listdir(wav_dir):
whole_filepath = os.path.join(wav_dir, file)
if os.path.isfile(whole_filepath) and str(whole_filepath).endswith(
".wav"):
wav_files.append(whole_filepath)
elif os.path.isdir(whole_filepath):
wav_files += get_wav_filelist(whole_filepath)
wav_files.sort()
return wav_files
# FILES SETUP:========================================================================
lu.mkdir(out_dir)
l_wavfiles = get_wav_filelist(wav_dir)
# MULTIPROCESSING EXTRACTION:==========================================================
lu.run_multithreaded(feat_extraction, l_wavfiles, out_dir)
# For debugging (don't delete):
#for wavfile in l_wavfiles:
# feat_extraction(wavfile, out_dir)
print('Done!')
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 b_config_merlin or b_wavgen:
# Edit Merlin's config file:
parser = configparser.ConfigParser()
parser.optionxform = str
parser.read([os.path.join(exper_path, 'conf/config_base.conf')])
parser['DEFAULT']['TOPLEVEL'] = exper_path
parser['Paths']['file_id_list'] = "%(work)s/" + file_id_list
parser['Labels']['question_file_name'] = os.path.join(
exper_path, question_file_name)
raise
except:
print2('Error while extracting features from',
path.join(wavdir, sentence + '.wav'), 'to', outdir)
flush2()
else:
print1(sentence)
flush1()
if __name__ == '__main__':
p = ArgumentParser()
p.add_argument('-s', '--senlst', dest='senlst', required=True)
p.add_argument('-c', '--config', dest='config', required=True)
p.add_argument('-d', '--debug', dest='debug', action='store_true')
a = p.parse_args()
load_config(a.config)
from __init__ import *
CONST_RATE = cfg_data.get('const', True)
with open(a.senlst) as f:
sentences = [l.rstrip() for l in f]
if a.debug:
for s in sentences:
extract(s, WAVDIR, ACO1DIR, CONST_RATE)
else:
lu.run_multithreaded(extract, sentences, WAVDIR, ACO1DIR, CONST_RATE)
# INPUT:==============================================================================
files_scp = '../demos/data_48k/file_id_predict.scp' # List of file names (tokens). Format used by Merlin.
in_feats_dir = '../demos/data_48k/params_predicted' # Input directory that contains the predicted features.
out_syn_dir = '../demos/data_48k/wavs_syn_from_predicted' # Where the synthesised waveform will be stored.
mag_dim = 60 # Number of Mel-scaled frequency bins.
phase_dim = 45 # Number of Mel-scaled frequency bins kept for phase features (real and imag). It must be <= mag_dim
pf_type = 'magphase' # "magphase": MagPhase's own postfilter (in development)
# "merlin": Merlin's style postfilter.
# "no": No postfilter.
b_multiproc = False # If True, it synthesises using all the available cores in parallel. If False, it just uses one core (slower).
# FILES SETUP:========================================================================
lu.mkdir(out_syn_dir)
l_file_tokns = lu.read_text_file2(files_scp, dtype='string',
comments='#').tolist()
# PROCESSING:=========================================================================
if b_multiproc:
lu.run_multithreaded(synthesis, in_feats_dir, l_file_tokns,
out_syn_dir, mag_dim, phase_dim, fs, pf_type)
else:
for file_tokn in l_file_tokns:
synthesis(in_feats_dir, file_tokn, out_syn_dir, mag_dim, phase_dim,
fs, pf_type)
print('Done!')
files_scp = '../data_48k/file_id.scp' # List of file names (tokens). Format used by Merlin.
in_feats_dir = '../data_48k/params' # Input directory that contains the predicted features.
out_syn_dir = '../data_48k/wavs_syn_merlin' # Where the synthesised waveform will be stored.
nbins_mel = 60 # Number of Mel-scaled frequency bins.
nbins_phase = 45 # Number of Mel-scaled frequency bins kept for phase features (real and imag). It must be <= nbins_mel
b_postfilter = True # If True, the MagPhase vocoder post-filter is applied. Note: If you want to use the one included in Merlin, disable this one.
b_parallel = False # If True, it synthesises using all the available cores in parallel. If False, it just uses one core (slower).
# FILES SETUP:========================================================================
lu.mkdir(out_syn_dir)
l_file_tokns = lu.read_text_file2(files_scp, dtype='string', comments='#').tolist()
# PROCESSING:=========================================================================
if b_parallel:
lu.run_multithreaded(synthesis, in_feats_dir, l_file_tokns, out_syn_dir, nbins_mel, nbins_phase, fs, fft_len, b_postfilter)
else:
for file_tokn in l_file_tokns:
synthesis(in_feats_dir, file_tokn, out_syn_dir, nbins_mel, nbins_phase, fs, fft_len, b_postfilter)
print('Done!')
fs = 48000
# INPUT:==============================================================================
files_scp = '../data_48k/file_id.scp' # List of file names (tokens). Format used by Merlin.
in_feats_dir = '../data_48k/params' # Input directory that contains the predicted features.
out_syn_dir = '../data_48k/wavs_syn_merlin' # Where the synthesised waveform will be stored.
nbins_mel = 60 # Number of Mel-scaled frequency bins.
nbins_phase = 45 # Number of Mel-scaled frequency bins kept for phase features (real and imag). It must be <= nbins_mel
b_postfilter = True # If True, the MagPhase vocoder post-filter is applied. Note: If you want to use the one included in Merlin, disable this one.
b_parallel = False # If True, it synthesises using all the available cores in parallel. If False, it just uses one core (slower).
# FILES SETUP:========================================================================
lu.mkdir(out_syn_dir)
l_file_tokns = lu.read_text_file2(files_scp, dtype='string', comments='#').tolist()
# PROCESSING:=========================================================================
if b_parallel:
lu.run_multithreaded(synthesis, in_feats_dir, l_file_tokns, out_syn_dir, nbins_mel, nbins_phase, fs, b_postfilter)
else:
for file_tokn in l_file_tokns:
synthesis(in_feats_dir, file_tokn, out_syn_dir, nbins_mel, nbins_phase, fs, b_postfilter)
print('Done!')
save_config(pars_acous_synth, join(acous_model_conf_path, 'acous_synth.conf'))
copy2(join(this_dir, 'conf_base', 'logging_config.conf'), join(exper_path, 'acoustic_model', 'conf', 'logging_config.conf'))
# Read file list:
file_id_list = pars_acous_train['Paths']['file_id_list']
l_file_tokns = lu.read_text_file2(file_id_list, dtype='string', comments='#').tolist()
acoustic_feats_path = pars_acous_train['Paths']['in_acous_feats_dir']
# Acoustic Feature Extraction:-------------------------------------------------------------
if b_feat_extr:
# Extract features:
lu.mkdir(acoustic_feats_path)
if b_feat_ext_multiproc:
lu.run_multithreaded(feat_extraction, join(exper_path, 'acoustic_model', 'data', 'wav'), l_file_tokns, acoustic_feats_path, d_mp_opts)
else:
for file_name_token in l_file_tokns:
feat_extraction(join(exper_path, 'acoustic_model', 'data', 'wav'), file_name_token, acoustic_feats_path, d_mp_opts)
# Labels Conversion to Variable Frame Rate:------------------------------------------------
if b_conv_labs_rate and not d_mp_opts['b_const_rate']: # NOTE: The script ./script/label_st_align_to_var_rate.py can be also called from comand line directly.
label_state_align = join(exper_path, 'acoustic_model', 'data', 'label_state_align')
label_state_align_var_rate = pars_acous_train['Labels']['label_align']
fs = int(pars_acous_train['Waveform']['samplerate'])
ltvr.convert(file_id_list,label_state_align, acoustic_feats_path, fs, label_state_align_var_rate)
# Run duration training:-------------------------------------------------------------------
if b_dur_train:
call([submit_path, run_merlin_path, join(dur_model_conf_path, 'dur_train.conf')])
return
def get_wav_filelist(wav_dir):
wav_files = []
for file in os.listdir(wav_dir):
whole_filepath = os.path.join(wav_dir, file)
if os.path.isfile(whole_filepath) and str(whole_filepath).endswith(".wav"):
wav_files.append(whole_filepath)
elif os.path.isdir(whole_filepath):
wav_files += get_wav_filelist(whole_filepath)
wav_files.sort()
return wav_files
# FILES SETUP:========================================================================
lu.mkdir(out_dir)
l_wavfiles = get_wav_filelist(wav_dir)
# MULTIPROCESSING EXTRACTION:==========================================================
lu.run_multithreaded(feat_extraction, l_wavfiles, out_dir)
# For debugging (don't delete):
#for wavfile in l_wavfiles:
# feat_extraction(wavfile, out_dir)
print('Done!')