def _analyze_world_features(self, x):
feat = FeatureExtractor(
analyzer="world",
fs=self.conf["fs"],
fftl=self.conf["fftl"],
shiftms=self.conf["shiftms"],
minf0=self.sconf["minf0"],
maxf0=self.sconf["maxf0"],
)
# analyze world based features
self.feats["f0"], self.feats["spc"], self.feats["ap"] = feat.analyze(x)
self.feats["mcep"] = feat.mcep(dim=self.conf["mcep_dim"],
alpha=self.conf["mcep_alpha"])
self.feats["npow"] = feat.npow()
self.feats["cap"] = feat.codeap()
cap = self.feats["cap"]
self.feats["ccap"] = np.zeros(cap.shape)
self.feats["cap_uv"] = np.zeros(cap.shape)
for d in range(self.feats["ccap"].shape[-1]):
cap[np.where(cap[:, d] == max(cap[:, d])), d] = 0.0
self.feats["cap_uv"][:, d], self.feats[
"ccap"][:, d] = convert_continuos_f0(cap[:, d])
self.feats["uv"], self.feats["cf0"] = convert_continuos_f0(
self.feats["f0"])
self.feats["lf0"] = np.log(self.feats["f0"] + EPS)
self.feats["lcf0"] = np.log(self.feats["cf0"])
def _analyze_world_features(self, x, f0_only=False):
feat = FeatureExtractor(
analyzer="world",
fs=self.conf["fs"],
fftl=self.conf["fftl"],
shiftms=self.conf["shiftms"],
minf0=self.sconf["minf0"],
maxf0=self.sconf["maxf0"],
)
# analyze world based features
self.feats["f0"], self.feats["spc"], self.feats["ap"] = feat.analyze(x)
self.feats["uv"], self.feats["cf0"] = convert_continuos_f0(
self.feats["f0"])
self.feats["lf0"] = np.log(self.feats["f0"] + EPS)
self.feats["lcf0"] = np.log(self.feats["cf0"])
if f0_only:
return
if self.conf["fftl"] != 256 and self.conf["fs"] > 16000:
# NOTE: 256 fft_size sometimes causes errors
self.feats["mcep"] = feat.mcep(dim=self.conf["mcep_dim"],
alpha=self.conf["mcep_alpha"])
self.feats["npow"] = feat.npow()
self.feats["cap"] = feat.codeap()
cap = self.feats["cap"]
self.feats["ccap"] = np.zeros(cap.shape)
self.feats["cap_uv"] = np.zeros(cap.shape)
for d in range(self.feats["ccap"].shape[-1]):
cap[np.where(cap[:, d] == max(cap[:, d])), d] = 0.0
(
self.feats["cap_uv"][:, d],
self.feats["ccap"][:, d],
) = convert_continuos_f0(cap[:, d])
def test_mod_power(self):
path = os.path.join(dirpath, 'data', 'test16000.wav')
fs, x = wavfile.read(path)
af = FeatureExtractor(analyzer='world', fs=fs, shiftms=5)
f0, _, ap = af.analyze(x)
mcep = af.mcep(dim=24, alpha=0.42)
rmcep = mcep
cvmcep = mcep * 1.50
modified_cvmcep = mod_power(cvmcep, rmcep, alpha=0.42)
assert modified_cvmcep.shape == cvmcep.shape
def test_anasyn_16000(self):
path = dirpath + '/data/test16000.wav'
fs, x = wavfile.read(path)
af = FeatureExtractor(analyzer='world', fs=fs, shiftms=5, fftl=1024)
f0, spc, ap = af.analyze(x)
mcep = af.mcep(dim=24, alpha=0.42)
assert len(np.nonzero(f0)[0]) > 0
assert spc.shape == ap.shape
# synthesize F0, mcep, ap
synth = Synthesizer(fs=fs, fftl=1024, shiftms=5)
wav = synth.synthesis(f0, mcep, ap, alpha=0.42)
nun_check(wav)
def test_anasyn_44100(self):
path = dirpath + '/data/test44100.wav'
fs, x = wavfile.read(path)
af = FeatureExtractor(analyzer='world', fs=fs, shiftms=5, minf0=100, fftl=2048)
f0, spc, ap = af.analyze(x)
mcep = af.mcep(dim=40, alpha=0.50)
assert len(np.nonzero(f0)[0]) > 0
assert spc.shape == ap.shape
# mcep synthesis
synth = Synthesizer(fs=fs, fftl=2048, shiftms=5)
wav = synth.synthesis(f0, mcep, ap, alpha=0.50)
nun_check(wav)
def test_anasyn_16000(self):
path = dirpath + '/data/test16000.wav'
fs, x = wavfile.read(path)
af = FeatureExtractor(analyzer='world', fs=fs, shiftms=5, fftl=1024)
f0, spc, ap = af.analyze(x)
mcep = af.mcep(dim=24, alpha=0.42)
assert len(np.nonzero(f0)[0]) > 0
assert spc.shape == ap.shape
# synthesize F0, mcep, ap
synth = Synthesizer(fs=fs, fftl=1024, shiftms=5)
wav = synth.synthesis(f0, mcep, ap, alpha=0.42)
nun_check(wav)
def get_world_features(wavpath, spk, conf, spkr_conf):
x, fs = sf.read(str(wavpath))
x = np.array(x, dtype=np.float)
x = low_cut_filter(x, fs, cutoff=70)
fe = FeatureExtractor(
analyzer="world",
fs=conf["feature"]["fs"],
fftl=conf["feature"]["fftl"],
shiftms=conf["feature"]["shiftms"],
minf0=spkr_conf[spk]["minf0"],
maxf0=spkr_conf[spk]["maxf0"],
)
cv_f0, _, _ = fe.analyze(x)
cv_mcep = fe.mcep(dim=conf["feature"]["mcep_dim"],
alpha=conf["feature"]["mcep_alpha"])
return cv_mcep, cv_f0
def test_anasyn_44100(self):
path = dirpath + '/data/test44100.wav'
fs, x = wavfile.read(path)
af = FeatureExtractor(analyzer='world',
fs=fs,
shiftms=5,
minf0=100,
fftl=2048)
f0, spc, ap = af.analyze(x)
mcep = af.mcep(dim=40, alpha=0.50)
assert len(np.nonzero(f0)[0]) > 0
assert spc.shape == ap.shape
# mcep synthesis
synth = Synthesizer(fs=fs, fftl=2048, shiftms=5)
wav = synth.synthesis(f0, mcep, ap, alpha=0.50)
nun_check(wav)
def main(*argv):
argv = argv if argv else sys.argv
# Options for python
dcp = 'Extract aoucstic features for the speaker'
parser = argparse.ArgumentParser(description=dcp)
parser.add_argument('--overwrite',
default=False,
action='store_true',
help='Overwrite h5 file')
parser.add_argument('speaker', type=str, help='Input speaker label')
parser.add_argument('ymlf', type=str, help='Yml file of the input speaker')
parser.add_argument('list_file',
type=str,
help='List file of the input speaker')
parser.add_argument('wav_dir',
type=str,
help='Wav file directory of the speaker')
parser.add_argument('pair_dir',
type=str,
help='Directory of the speaker pair')
args = parser.parse_args(argv)
# read parameters from speaker yml
sconf = SpeakerYML(args.ymlf)
h5_dir = os.path.join(args.pair_dir, 'h5')
anasyn_dir = os.path.join(args.pair_dir, 'anasyn')
if not os.path.exists(os.path.join(h5_dir, args.speaker)):
os.makedirs(os.path.join(h5_dir, args.speaker))
if not os.path.exists(os.path.join(anasyn_dir, args.speaker)):
os.makedirs(os.path.join(anasyn_dir, args.speaker))
# constract FeatureExtractor class
feat = FeatureExtractor(analyzer=sconf.analyzer,
fs=sconf.wav_fs,
fftl=sconf.wav_fftl,
shiftms=sconf.wav_shiftms,
minf0=sconf.f0_minf0,
maxf0=sconf.f0_maxf0)
# constract Synthesizer class
synthesizer = Synthesizer(fs=sconf.wav_fs,
fftl=sconf.wav_fftl,
shiftms=sconf.wav_shiftms)
# open list file
with open(args.list_file, 'r') as fp:
for line in fp:
f = line.rstrip()
h5f = os.path.join(h5_dir, f + '.h5')
if (not os.path.exists(h5f)) or args.overwrite:
wavf = os.path.join(args.wav_dir, f + '.wav')
fs, x = wavfile.read(wavf)
x = np.array(x, dtype=np.float)
x = low_cut_filter(x, fs, cutoff=70)
assert fs == sconf.wav_fs
print("Extract acoustic features: " + wavf)
# analyze F0, spc, and ap
f0, spc, ap = feat.analyze(x)
mcep = feat.mcep(dim=sconf.mcep_dim, alpha=sconf.mcep_alpha)
npow = feat.npow()
codeap = feat.codeap()
# save features into a hdf5 file
h5 = HDF5(h5f, mode='w')
h5.save(f0, ext='f0')
# h5.save(spc, ext='spc')
# h5.save(ap, ext='ap')
h5.save(mcep, ext='mcep')
h5.save(npow, ext='npow')
h5.save(codeap, ext='codeap')
h5.close()
# analysis/synthesis using F0, mcep, and ap
wav = synthesizer.synthesis(
f0,
mcep,
ap,
alpha=sconf.mcep_alpha,
)
wav = np.clip(wav, -32768, 32767)
anasynf = os.path.join(anasyn_dir, f + '.wav')
wavfile.write(anasynf, fs, np.array(wav, dtype=np.int16))
else:
print("Acoustic features already exist: " + h5f)
def main(*argv):
argv = argv if argv else sys.argv[1:]
# Options for python
description = 'estimate joint feature of source and target speakers'
parser = argparse.ArgumentParser(description=description)
parser.add_argument('-gmmmode',
'--gmmmode',
type=str,
default=None,
help='mode of the GMM [None, diff, or intra]')
parser.add_argument('org', type=str, help='Original speaker')
parser.add_argument('tar', type=str, help='Target speaker')
parser.add_argument('org_yml',
type=str,
help='Yml file of the original speaker')
parser.add_argument('pair_yml',
type=str,
help='Yml file of the speaker pair')
parser.add_argument('eval_list_file',
type=str,
help='List file for evaluation')
parser.add_argument('wav_dir',
type=str,
help='Directory path of source spekaer')
parser.add_argument('pair_dir',
type=str,
help='Directory path of pair directory')
args = parser.parse_args(argv)
# read parameters from speaker yml
sconf = SpeakerYML(args.org_yml)
pconf = PairYML(args.pair_yml)
# read GMM for mcep
mcepgmmpath = os.path.join(args.pair_dir, 'model/GMM_mcep.pkl')
mcepgmm = GMMConvertor(
n_mix=pconf.GMM_mcep_n_mix,
covtype=pconf.GMM_mcep_covtype,
gmmmode=args.gmmmode,
)
param = joblib.load(mcepgmmpath)
mcepgmm.open_from_param(param)
print("GMM for mcep conversion mode: {}".format(args.gmmmode))
# read F0 statistics
stats_dir = os.path.join(args.pair_dir, 'stats')
orgstatspath = os.path.join(stats_dir, args.org + '.h5')
orgstats_h5 = HDF5(orgstatspath, mode='r')
orgf0stats = orgstats_h5.read(ext='f0stats')
orgstats_h5.close()
# read F0 and GV statistics for target
tarstatspath = os.path.join(stats_dir, args.tar + '.h5')
tarstats_h5 = HDF5(tarstatspath, mode='r')
tarf0stats = tarstats_h5.read(ext='f0stats')
targvstats = tarstats_h5.read(ext='gv')
tarstats_h5.close()
# read GV statistics for converted mcep
cvgvstatspath = os.path.join(args.pair_dir, 'model', 'cvgv.h5')
cvgvstats_h5 = HDF5(cvgvstatspath, mode='r')
cvgvstats = cvgvstats_h5.read(ext='cvgv')
diffcvgvstats = cvgvstats_h5.read(ext='diffcvgv')
cvgvstats_h5.close()
mcepgv = GV()
f0stats = F0statistics()
# constract FeatureExtractor class
feat = FeatureExtractor(analyzer=sconf.analyzer,
fs=sconf.wav_fs,
fftl=sconf.wav_fftl,
shiftms=sconf.wav_shiftms,
minf0=sconf.f0_minf0,
maxf0=sconf.f0_maxf0)
# constract Synthesizer class
synthesizer = Synthesizer(fs=sconf.wav_fs,
fftl=sconf.wav_fftl,
shiftms=sconf.wav_shiftms)
# test directory
test_dir = os.path.join(args.pair_dir, 'test')
os.makedirs(os.path.join(test_dir, args.org), exist_ok=True)
# conversion in each evaluation file
with open(args.eval_list_file, 'r') as fp:
for line in fp:
# open wav file
f = line.rstrip()
wavf = os.path.join(args.wav_dir, f + '.wav')
fs, x = wavfile.read(wavf)
x = x.astype(np.float)
x = low_cut_filter(x, fs, cutoff=70)
assert fs == sconf.wav_fs
# analyze F0, mcep, and ap
f0, spc, ap = feat.analyze(x)
mcep = feat.mcep(dim=sconf.mcep_dim, alpha=sconf.mcep_alpha)
mcep_0th = mcep[:, 0]
# convert F0
cvf0 = f0stats.convert(f0, orgf0stats, tarf0stats)
# convert mcep
cvmcep_wopow = mcepgmm.convert(static_delta(mcep[:, 1:]),
cvtype=pconf.GMM_mcep_cvtype)
cvmcep = np.c_[mcep_0th, cvmcep_wopow]
# synthesis VC w/ GV
if args.gmmmode is None:
cvmcep_wGV = mcepgv.postfilter(cvmcep,
targvstats,
cvgvstats=cvgvstats,
alpha=pconf.GV_morph_coeff,
startdim=1)
wav = synthesizer.synthesis(
cvf0,
cvmcep_wGV,
ap,
rmcep=mcep,
alpha=sconf.mcep_alpha,
)
wavpath = os.path.join(test_dir, f + '_VC.wav')
# synthesis DIFFVC w/ GV
if args.gmmmode == 'diff':
cvmcep[:, 0] = 0.0
cvmcep_wGV = mcepgv.postfilter(mcep + cvmcep,
targvstats,
cvgvstats=diffcvgvstats,
alpha=pconf.GV_morph_coeff,
startdim=1) - mcep
wav = synthesizer.synthesis_diff(
x,
cvmcep_wGV,
rmcep=mcep,
alpha=sconf.mcep_alpha,
)
wavpath = os.path.join(test_dir, f + '_DIFFVC.wav')
# write waveform
if not os.path.exists(os.path.join(test_dir, f)):
os.makedirs(os.path.join(test_dir, f))
wav = np.clip(wav, -32768, 32767)
wavfile.write(wavpath, fs, wav.astype(np.int16))
print(wavpath)
def world_feature_extract(queue, wav_list, config):
"""WORLD feature extraction
Args:
queue (multiprocessing.Queue): the queue to store the file name of utterance
wav_list (list): list of the wav files
config (dict): feature extraction config
"""
# define feature extractor
feature_extractor = FeatureExtractor(analyzer="world",
fs=config['sampling_rate'],
shiftms=config['shiftms'],
minf0=config['minf0'],
maxf0=config['maxf0'],
fftl=config['fft_size'])
# extraction
for i, wav_name in enumerate(wav_list):
logging.info("now processing %s (%d/%d)" %
(wav_name, i + 1, len(wav_list)))
# load wavfile and apply low cut filter
fs, x = wavfile.read(wav_name)
x = np.array(x, dtype=np.float32)
if config['highpass_cutoff'] != 0:
x = low_cut_filter(x, fs, cutoff=config['highpass_cutoff'])
# check sampling frequency
if not fs == config['sampling_rate']:
logging.error("sampling frequency of %s is not matched." %
wav_name)
sys.exit(1)
# extract features
f0, spc, ap = feature_extractor.analyze(x)
codeap = feature_extractor.codeap()
mcep = feature_extractor.mcep(dim=config['mcep_dim'],
alpha=config['mcep_alpha'])
npow = feature_extractor.npow()
uv, cont_f0 = convert_continuos_f0(f0)
lpf_fs = int(1.0 / (config['shiftms'] * 0.001))
cont_f0_lpf = low_pass_filter(cont_f0, lpf_fs, cutoff=20)
next_cutoff = 70
while not (cont_f0_lpf >= [0]).all():
cont_f0_lpf = low_pass_filter(cont_f0, lpf_fs, cutoff=next_cutoff)
next_cutoff *= 2
# concatenate
cont_f0_lpf = np.expand_dims(cont_f0_lpf, axis=-1)
uv = np.expand_dims(uv, axis=-1)
feats = np.concatenate([uv, cont_f0_lpf, mcep, codeap], axis=1)
# save feature
feat_name = path_replace(wav_name,
config['indir'],
config['outdir'],
extname=config['feature_format'])
write_hdf5(feat_name, "/%s" % (config["feat_type"]), feats)
if config['save_f0']:
write_hdf5(feat_name, "/f0", f0)
if config['save_ap']:
write_hdf5(feat_name, "/ap", ap)
if config['save_spc']:
write_hdf5(feat_name, "/spc", spc)
if config['save_npow']:
write_hdf5(feat_name, "/npow", npow)
queue.put('Finish')
def main(*argv):
argv = argv if argv else sys.argv
# Options for python
dcp = 'Extract aoucstic features for the speaker'
parser = argparse.ArgumentParser(description=dcp)
parser.add_argument('--overwrite', default=False, action='store_true',
help='Overwrite h5 file')
parser.add_argument('speaker', type=str,
help='Input speaker label')
parser.add_argument('ymlf', type=str,
help='Yml file of the input speaker')
parser.add_argument('list_file', type=str,
help='List file of the input speaker')
parser.add_argument('wav_dir', type=str,
help='Wav file directory of the speaker')
parser.add_argument('pair_dir', type=str,
help='Directory of the speaker pair')
args = parser.parse_args(argv)
# read parameters from speaker yml
sconf = SpeakerYML(args.ymlf)
h5_dir = os.path.join(args.pair_dir, 'h5')
anasyn_dir = os.path.join(args.pair_dir, 'anasyn')
if not os.path.exists(os.path.join(h5_dir, args.speaker)):
os.makedirs(os.path.join(h5_dir, args.speaker))
if not os.path.exists(os.path.join(anasyn_dir, args.speaker)):
os.makedirs(os.path.join(anasyn_dir, args.speaker))
# constract FeatureExtractor class
feat = FeatureExtractor(analyzer=sconf.analyzer,
fs=sconf.wav_fs,
fftl=sconf.wav_fftl,
shiftms=sconf.wav_shiftms,
minf0=sconf.f0_minf0,
maxf0=sconf.f0_maxf0)
# constract Synthesizer class
synthesizer = Synthesizer(fs=sconf.wav_fs,
fftl=sconf.wav_fftl,
shiftms=sconf.wav_shiftms)
# open list file
with open(args.list_file, 'r') as fp:
for line in fp:
f = line.rstrip()
h5f = os.path.join(h5_dir, f + '.h5')
if (not os.path.exists(h5f)) or args.overwrite:
wavf = os.path.join(args.wav_dir, f + '.wav')
fs, x = wavfile.read(wavf)
x = np.array(x, dtype=np.float)
x = low_cut_filter(x, fs, cutoff=70)
assert fs == sconf.wav_fs
print("Extract acoustic features: " + wavf)
# analyze F0, spc, and ap
f0, spc, ap = feat.analyze(x)
mcep = feat.mcep(dim=sconf.mcep_dim, alpha=sconf.mcep_alpha)
npow = feat.npow()
codeap = feat.codeap()
# save features into a hdf5 file
h5 = HDF5(h5f, mode='w')
h5.save(f0, ext='f0')
# h5.save(spc, ext='spc')
# h5.save(ap, ext='ap')
h5.save(mcep, ext='mcep')
h5.save(npow, ext='npow')
h5.save(codeap, ext='codeap')
h5.close()
# analysis/synthesis using F0, mcep, and ap
wav = synthesizer.synthesis(f0,
mcep,
ap,
alpha=sconf.mcep_alpha,
)
wav = np.clip(wav, -32768, 32767)
anasynf = os.path.join(anasyn_dir, f + '.wav')
wavfile.write(anasynf, fs, np.array(wav, dtype=np.int16))
else:
print("Acoustic features already exist: " + h5f)
def main():
if args.file == 'con':
file = 'converted'
elif args.file == 'tar':
file = 'target'
else:
raise ValueError("The file is incorrect")
feat = FeatureExtractor(analyzer='world',
fs=22050,
fftl=1024,
shiftms=5,
minf0=args.minf0,
maxf0=args.maxf0)
# constract Synthesizer class
synthesizer = Synthesizer(fs=22050, fftl=1024, shiftms=5)
# open list file
with open('./list/' + file + '.list', 'r') as fp:
for line in fp:
f = line.rstrip()
h5f = os.path.join('./' + file + '/h5f/', f + '.h5')
if (not os.path.exists(h5f)):
wavf = os.path.join('./' + file + '/wav/', f + '.wav')
fs, x = wavfile.read(wavf)
x = np.array(x, dtype=np.float)
x = low_cut_filter(x, fs, cutoff=70)
print("Extract acoustic features: " + wavf)
# analyze F0, spc, and ap
f0, spc, ap = feat.analyze(x)
mcep = feat.mcep(dim=34, alpha=0.544)
npow = feat.npow()
codeap = feat.codeap()
# save features into a hdf5 file
h5 = HDF5(h5f, mode='w')
h5.save(f0, ext='f0')
# h5.save(spc, ext='spc')
# h5.save(ap, ext='ap')
h5.save(mcep, ext='mcep')
h5.save(npow, ext='npow')
h5.save(codeap, ext='codeap')
h5.close()
# analysis/synthesis using F0, mcep, and ap
wav = synthesizer.synthesis(
f0,
mcep,
ap,
alpha=0.544,
)
wav = np.clip(wav, -32768, 32767)
anasynf = os.path.join('./' + file + '/anasyn/', f + '.wav')
wavfile.write(anasynf, fs, np.array(wav, dtype=np.int16))
else:
print("Acoustic features already exist: " + h5f)
