def validate_features_dataset(output_dataset_path, ds_validation_path):
ds = F.Dataset(output_dataset_path, read_only=True)
print(ds)
features = {}
for key, val in ds.items():
if 'indices_' in key:
name = key.split('_')[-1]
features[name] = (val, ds[name])
all_indices = [val[0] for val in features.values()]
# ====== sampling 250 files ====== #
all_files = sampling_iter(it=all_indices[0].keys(),
k=250,
seed=Config.SUPER_SEED)
all_files = [f for f in all_files if all(f in ids for ids in all_indices)]
print("#Samples:", ctext(len(all_files), 'cyan'))
# ====== ignore the 20-figures warning ====== #
with catch_warnings_ignore(RuntimeWarning):
for file_name in all_files:
X = {}
for feat_name, (ids, data) in features.items():
start, end = ids[file_name]
X[feat_name] = data[start:end][:].astype('float32')
V.plot_multiple_features(features=X,
fig_width=20,
title='[%s]%s' %
(ds['dsname'][file_name], file_name))
V.plot_save(ds_validation_path, dpi=12)
def validate_features_dataset(output_dataset_path, ds_validation_path):
ds = F.Dataset(output_dataset_path, read_only=True)
print(ds)
features = {}
for key, val in ds.items():
if 'indices_' in key:
name = key.split('_')[-1]
features[name] = (val, ds[name])
all_indices = [val[0] for val in features.values()]
# ====== sampling 250 files ====== #
all_files = sampling_iter(it=all_indices[0].keys(), k=250,
seed=Config.SUPER_SEED)
all_files = [f for f in all_files
if all(f in ids for ids in all_indices)]
print("#Samples:", ctext(len(all_files), 'cyan'))
# ====== ignore the 20-figures warning ====== #
with catch_warnings_ignore(RuntimeWarning):
for file_name in all_files:
X = {}
for feat_name, (ids, data) in features.items():
start, end = ids[file_name]
X[feat_name] = data[start:end][:].astype('float32')
V.plot_multiple_features(features=X, fig_width=20,
title='[%s]%s' % (ds['dsname'][file_name], file_name))
V.plot_save(ds_validation_path, dpi=12)
def visualize_latent_space(X_org, X_latent, name, labels, title):
"""
X_org : [n_samples, n_timesteps, n_features]
X_latent : [n_samples, n_timesteps, n_latents]
"""
assert X_org.shape[0] == X_latent.shape[0] == len(name) == len(labels)
assert not np.any(np.isnan(X_org))
assert not np.any(np.isnan(X_latent))
X_org = X_org.astype('float32')
X_latent = X_latent.astype('float32')
# ====== evaluation of the latent space ====== #
n_channels = 1 if X_latent.ndim == 3 else int(np.prod(X_latent.shape[3:]))
n_samples = X_org.shape[0]
# 1 for original, 1 for mean channel, then the rest
n_row = 1 + 1 + n_channels
n_col = 3
V.plot_figure(nrow=n_row + 1, ncol=16)
# only select 3 random sample
for i, idx in enumerate(
sampling_iter(it=range(n_samples), k=n_col, seed=1234)):
x = X_org[idx]
# latent tensor can be 3D or 4D
z = X_latent[idx]
if z.ndim > 3:
z = np.reshape(z, newshape=(z.shape[0], z.shape[1], -1))
elif z.ndim == 2:
z = np.reshape(z, newshape=(z.shape[0], z.shape[1], 1))
elif z.ndim == 3:
pass
else:
raise ValueError("No support for z value: %s" % str(z.shape))
# plot original acoustic
ax = V.plot_spectrogram(x.T, ax=(n_row, n_col, i + 1), title='Org')
if i == 0:
ax.set_title("[%s]'%s-%s'" %
(str(title), str(name[idx]), str(labels[idx])),
fontsize=8)
else:
ax.set_title("'%s-%s'" % (str(name[idx]), str(labels[idx])),
fontsize=8)
# plot the mean
V.plot_spectrogram(np.mean(z, axis=-1).T,
ax=(n_row, n_col, i + 4),
title='Zmean')
# plot first 25 channels
if n_channels > 1:
for j in range(min(8, n_channels)):
V.plot_spectrogram(z[:, :, j].T,
ax=(n_row, n_col, j * 3 + 7 + i),
title='Z%d' % j)
def visualize_latent_space(X_org, X_latent, name, labels, title):
"""
X_org : [n_samples, n_timesteps, n_features]
X_latent : [n_samples, n_timesteps, n_latents]
"""
assert X_org.shape[0] == X_latent.shape[0] == len(name) == len(labels)
assert not np.any(np.isnan(X_org))
assert not np.any(np.isnan(X_latent))
X_org = X_org.astype('float32')
X_latent = X_latent.astype('float32')
# ====== evaluation of the latent space ====== #
n_channels = 1 if X_latent.ndim == 3 else int(np.prod(X_latent.shape[3:]))
n_samples = X_org.shape[0]
# 1 for original, 1 for mean channel, then the rest
n_row = 1 + 1 + n_channels
n_col = 3
V.plot_figure(nrow=n_row + 1, ncol=16)
# only select 3 random sample
for i, idx in enumerate(
sampling_iter(it=range(n_samples), k= n_col, seed=5218)):
x = X_org[idx]
# latent tensor can be 3D or 4D
z = X_latent[idx]
if z.ndim > 3:
z = np.reshape(z, newshape=(z.shape[0], z.shape[1], -1))
elif z.ndim == 2:
z = np.reshape(z, newshape=(z.shape[0], z.shape[1], 1))
elif z.ndim == 3:
pass
else:
raise ValueError("No support for z value: %s" % str(z.shape))
# plot original acoustic
ax = V.plot_spectrogram(x.T, ax=(n_row, n_col, i + 1), title='Org')
if i == 0:
ax.set_title("[%s]'%s-%s'" % (str(title), str(name[idx]), str(labels[idx])),
fontsize=8)
else:
ax.set_title("'%s-%s'" % (str(name[idx]), str(labels[idx])),
fontsize=8)
# plot the mean
V.plot_spectrogram(np.mean(z, axis=-1).T,
ax=(n_row, n_col, i + 4), title='Zmean')
# plot first 25 channels
if n_channels > 1:
for j in range(min(8, n_channels)):
V.plot_spectrogram(z[:, :, j].T,
ax=(n_row, n_col, j * 3 + 7 + i),
title='Z%d' % j)
def prepare_dnn_data(recipe, feat, utt_length, seed=87654321):
"""
Return
------
train_feeder : Feeder for training
valid_feeder : Feeder for validating
test_ids : Test indices
test_dat : Data array
all_speakers : list of all speaker in training set
"""
# Load dataset
frame_length = int(utt_length / FRAME_SHIFT)
ds = F.Dataset(os.path.join(PATH_ACOUSTIC_FEAT, recipe), read_only=True)
X = ds[feat]
train_indices = {name: ds['indices'][name] for name in TRAIN_DATA.keys()}
test_indices = {
name: start_end
for name, start_end in ds['indices'].items() if name not in TRAIN_DATA
}
train_indices, valid_indices = train_valid_test_split(x=list(
train_indices.items()),
train=0.9,
inc_test=False,
seed=seed)
all_speakers = sorted(set(TRAIN_DATA.values()))
n_speakers = max(all_speakers) + 1
print("#Train files:", ctext(len(train_indices), 'cyan'))
print("#Valid files:", ctext(len(valid_indices), 'cyan'))
print("#Test files:", ctext(len(test_indices), 'cyan'))
print("#Speakers:", ctext(n_speakers, 'cyan'))
recipes = [
F.recipes.Sequencing(frame_length=frame_length,
step_length=frame_length,
end='pad',
pad_value=0,
pad_mode='post',
data_idx=0),
F.recipes.Name2Label(lambda name: TRAIN_DATA[name], ref_idx=0),
F.recipes.LabelOneHot(nb_classes=n_speakers, data_idx=1)
]
train_feeder = F.Feeder(data_desc=F.IndexedData(data=X,
indices=train_indices),
batch_mode='batch',
ncpu=7,
buffer_size=12)
valid_feeder = F.Feeder(data_desc=F.IndexedData(data=X,
indices=valid_indices),
batch_mode='batch',
ncpu=2,
buffer_size=4)
train_feeder.set_recipes(recipes)
valid_feeder.set_recipes(recipes)
print(train_feeder)
# ====== cache the test data ====== #
cache_dat = os.path.join(PATH_EXP,
'test_%s_%d.dat' % (feat, int(utt_length)))
cache_ids = os.path.join(PATH_EXP,
'test_%s_%d.ids' % (feat, int(utt_length)))
# validate cache files
if os.path.exists(cache_ids):
with open(cache_ids, 'rb') as f:
ids = pickle.load(f)
if len(ids) != len(test_indices):
os.remove(cache_ids)
if os.path.exists(cache_dat):
os.remove(cache_dat)
elif os.path.exists(cache_dat):
os.remove(cache_dat)
# caching
if not os.path.exists(cache_dat):
dat = F.MmapData(cache_dat,
dtype='float16',
shape=(0, frame_length, X.shape[1]))
ids = {}
prog = Progbar(target=len(test_indices))
s = 0
for name, (start, end) in test_indices.items():
y = X[start:end]
y = segment_axis(y,
axis=0,
frame_length=frame_length,
step_length=frame_length,
end='pad',
pad_value=0,
pad_mode='post')
dat.append(y)
# update indices
ids[name] = (s, s + len(y))
s += len(y)
# update progress
prog.add(1)
dat.flush()
dat.close()
with open(cache_ids, 'wb') as f:
pickle.dump(ids, f)
# ====== re-load ====== #
dat = F.MmapData(cache_dat, read_only=True)
with open(cache_ids, 'rb') as f:
ids = pickle.load(f)
# ====== save some sample ====== #
sample_path = os.path.join(PATH_EXP,
'test_%s_%d.pdf' % (feat, int(utt_length)))
V.plot_figure(nrow=9, ncol=6)
for i, (name, (start, end)) in enumerate(
sampling_iter(it=sorted(ids.items(), key=lambda x: x[0]),
k=12,
seed=87654321)):
x = dat[start:end][:].astype('float32')
ax = V.plot_spectrogram(x[np.random.randint(0, len(x))].T,
ax=(12, 1, i + 1),
title='')
ax.set_title(name)
V.plot_save(sample_path)
return (train_feeder, valid_feeder, ids, dat, all_speakers)
os.mkdir(PATH_ACOUSTIC_FEAT)
# ====== remove '_quarter' if you want full training data ====== #
FILE_LIST = "voxceleb_files_quarter"
TRAIN_LIST = "voxceleb_sys_train_with_labels_quarter"
TRIAL_LIST = "voxceleb_trials"
# ====== Load the file list ====== #
ds = F.load_voxceleb_list()
WAV_FILES = {} # dictionary mapping 'file_path' -> 'file_name'
for path, channel, name in ds[FILE_LIST]:
path = os.path.join(PATH_TO_WAV, path)
# validate all files are exist
assert os.path.exists(path), path
WAV_FILES[path] = name
# some sampled files for testing
SAMPLED_WAV_FILE = sampling_iter(it=sorted(WAV_FILES.items(),
key=lambda x: x[0]),
k=8,
seed=87654321)
# ====== extract the list of all train files ====== #
# mapping from name of training file to speaker label
TRAIN_DATA = {}
for x, y in ds[TRAIN_LIST]:
TRAIN_DATA[x] = int(y)
# ===========================================================================
# Path helpers
# ===========================================================================
def get_model_path(system_name, args):
"""Return: exp_dir, model_path, log_path, train_path, test_path"""
name = '_'.join([str(system_name).lower(), args.recipe, args.feat])
if 'l' in args:
def prepare_dnn_data(recipe, feat, utt_length, seed=52181208):
"""
Return
------
train_feeder : Feeder for training
valid_feeder : Feeder for validating
test_ids : Test indices
test_dat : Data array
all_speakers : list of all speaker in training set
"""
# Load dataset
frame_length = int(utt_length / FRAME_SHIFT)
ds = F.Dataset(os.path.join(PATH_ACOUSTIC_FEAT, recipe),
read_only=True)
X = ds[feat]
train_indices = {name: ds['indices'][name]
for name in TRAIN_DATA.keys()}
test_indices = {name: start_end
for name, start_end in ds['indices'].items()
if name not in TRAIN_DATA}
train_indices, valid_indices = train_valid_test_split(
x=list(train_indices.items()), train=0.9, inc_test=False, seed=seed)
all_speakers = sorted(set(TRAIN_DATA.values()))
n_speakers = max(all_speakers) + 1
print("#Train files:", ctext(len(train_indices), 'cyan'))
print("#Valid files:", ctext(len(valid_indices), 'cyan'))
print("#Test files:", ctext(len(test_indices), 'cyan'))
print("#Speakers:", ctext(n_speakers, 'cyan'))
recipes = [
F.recipes.Sequencing(frame_length=frame_length, step_length=frame_length,
end='pad', pad_value=0, pad_mode='post',
data_idx=0),
F.recipes.Name2Label(lambda name:TRAIN_DATA[name], ref_idx=0),
F.recipes.LabelOneHot(nb_classes=n_speakers, data_idx=1)
]
train_feeder = F.Feeder(
data_desc=F.IndexedData(data=X, indices=train_indices),
batch_mode='batch', ncpu=7, buffer_size=12)
valid_feeder = F.Feeder(
data_desc=F.IndexedData(data=X, indices=valid_indices),
batch_mode='batch', ncpu=2, buffer_size=4)
train_feeder.set_recipes(recipes)
valid_feeder.set_recipes(recipes)
print(train_feeder)
# ====== cache the test data ====== #
cache_dat = os.path.join(PATH_EXP, 'test_%s_%d.dat' % (feat, int(utt_length)))
cache_ids = os.path.join(PATH_EXP, 'test_%s_%d.ids' % (feat, int(utt_length)))
# validate cache files
if os.path.exists(cache_ids):
with open(cache_ids, 'rb') as f:
ids = pickle.load(f)
if len(ids) != len(test_indices):
os.remove(cache_ids)
if os.path.exists(cache_dat):
os.remove(cache_dat)
elif os.path.exists(cache_dat):
os.remove(cache_dat)
# caching
if not os.path.exists(cache_dat):
dat = F.MmapData(cache_dat, dtype='float16',
shape=(0, frame_length, X.shape[1]))
ids = {}
prog = Progbar(target=len(test_indices))
s = 0
for name, (start, end) in test_indices.items():
y = X[start:end]
y = segment_axis(y, axis=0,
frame_length=frame_length, step_length=frame_length,
end='pad', pad_value=0, pad_mode='post')
dat.append(y)
# update indices
ids[name] = (s, s + len(y))
s += len(y)
# update progress
prog.add(1)
dat.flush()
dat.close()
with open(cache_ids, 'wb') as f:
pickle.dump(ids, f)
# ====== re-load ====== #
dat = F.MmapData(cache_dat, read_only=True)
with open(cache_ids, 'rb') as f:
ids = pickle.load(f)
# ====== save some sample ====== #
sample_path = os.path.join(PATH_EXP,
'test_%s_%d.pdf' % (feat, int(utt_length)))
V.plot_figure(nrow=9, ncol=6)
for i, (name, (start, end)) in enumerate(
sampling_iter(it=sorted(ids.items(), key=lambda x: x[0]), k=12, seed=52181208)):
x = dat[start:end][:].astype('float32')
ax = V.plot_spectrogram(x[np.random.randint(0, len(x))].T,
ax=(12, 1, i + 1), title='')
ax.set_title(name)
V.plot_save(sample_path)
return (train_feeder, valid_feeder,
ids, dat, all_speakers)
os.mkdir(PATH_ACOUSTIC_FEAT)
# ====== remove '_quarter' if you want full training data ====== #
FILE_LIST = "voxceleb_files_quarter"
TRAIN_LIST = "voxceleb_sys_train_with_labels_quarter"
TRIAL_LIST = "voxceleb_trials"
# ====== Load the file list ====== #
ds = F.load_voxceleb_list()
WAV_FILES = {} # dictionary mapping 'file_path' -> 'file_name'
for path, channel, name in ds[FILE_LIST]:
path = os.path.join(PATH_TO_WAV, path)
# validate all files are exist
assert os.path.exists(path), path
WAV_FILES[path] = name
# some sampled files for testing
SAMPLED_WAV_FILE = sampling_iter(it=sorted(WAV_FILES.items(),
key=lambda x: x[0]),
k=8, seed=52181208)
# ====== extract the list of all train files ====== #
# mapping from name of training file to speaker label
TRAIN_DATA = {}
for x, y in ds[TRAIN_LIST]:
TRAIN_DATA[x] = int(y)
# ===========================================================================
# Path helpers
# ===========================================================================
def get_model_path(system_name, args):
"""Return: exp_dir, model_path, log_path, train_path, test_path"""
name = '_'.join([str(system_name).lower(), args.recipe, args.feat])
if 'l' in args:
name += '_' + str(int(args.l))