Skip to content

Commit

Permalink
Version 1.1.0 of tfglib
Browse files Browse the repository at this point in the history
  • Loading branch information
@albertaparicio
albertaparicio committed Jan 29, 2017
1 parent e112a04 commit a086125
Show file tree
Hide file tree
Showing 2 changed files with 33 additions and 21 deletions.
7 changes: 4 additions & 3 deletions setup.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,8 @@
"""

# Always prefer setuptools over distutils
from setuptools import setup, find_packages
from setuptools import setup
# from setuptools import find_packages
# To use a consistent encoding
from codecs import open
from os import path
Expand All @@ -23,7 +24,7 @@
# Versions should comply with PEP440. For a discussion on single-sourcing
# the version across setup.py and the project code, see
# https://packaging.python.org/en/latest/single_source_version.html
version='1.0.0',
version='1.1.0',

description='Python libraries for the Voice Conversion TFG',
long_description=long_description,
Expand Down Expand Up @@ -51,7 +52,7 @@
'Topic :: Software Development :: Build Tools',

# Pick your license as you wish (should match "license" above)
'License :: OSI Approved :: GNU Lesser General Public License v3 or later (LGPLv3+)',
'License :: OSI Approved :: GNU Lesser General Public License v3 or later (LGPLv3+)'

# Specify the Python versions you support here. In particular, ensure
# that you indicate whether you support Python 2, Python 3 or both.
Expand Down
47 changes: 29 additions & 18 deletions tfglib/seq2seq_datatable.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -80,14 +80,14 @@ def seq2seq_build_file_table(

# Check that source_dir and target_dir end in a '/'
try:
assert source_dir[len(source_dir) - 1] == '/'
assert source_dir[-1] == '/'
except AssertionError:
print(
"Please make sure the source data directory string ends with a '/'"
)

try:
assert target_dir[len(target_dir) - 1] == '/'
assert target_dir[-1] == '/'
except AssertionError:
print(
"Please make sure the target data directory string ends with a '/'"
Expand Down Expand Up @@ -138,10 +138,10 @@ def seq2seq_build_file_table(

# Initialize End-Of-Sequence flag
src_eos_flag = np.zeros(source_vf.shape)
src_eos_flag[src_eos_flag.shape[0] - 1, :] = 1
src_eos_flag[-1, :] = 1

trg_eos_flag = np.zeros(target_vf.shape)
trg_eos_flag[trg_eos_flag.shape[0] - 1, :] = 1
trg_eos_flag[-1, :] = 1

# Initialize one-hot-encoded speaker indexes
src_spk_index = to_categorical(
Expand Down Expand Up @@ -220,7 +220,7 @@ def seq2seq_construct_datatable(data_dir, speakers_file, basenames_file):

# Check that data_dir ends in a '/'
try:
assert data_dir[len(data_dir) - 1] == '/'
assert data_dir[-1] == '/'
except AssertionError:
print("Please, make sure the data directory string ends with a '/'")

Expand All @@ -240,8 +240,8 @@ def seq2seq_construct_datatable(data_dir, speakers_file, basenames_file):

# Initialize datatables
src_datatable = []
src_masks = []
trg_datatable = []
src_masks = []
trg_masks = []

# Initialize maximum and minimum values matrices
Expand All @@ -252,9 +252,7 @@ def seq2seq_construct_datatable(data_dir, speakers_file, basenames_file):
for src_index, src_spk in enumerate(speakers):
for trg_index, trg_spk in enumerate(speakers):
for basename in basenames:
print(src_spk)
print(trg_spk)
print(basename)
print(src_spk + '->' + trg_spk + ' ' + basename)

(aux_src_params,
aux_src_mask,
Expand All @@ -269,17 +267,24 @@ def seq2seq_construct_datatable(data_dir, speakers_file, basenames_file):
longest_seq
)

# Obtain maximum and minimum values of each parameter of each speaker
# Obtain maximum and minimum values of each speaker's parameter
# Mask parameters to avoid the zero-padded values
masked_params = np.ma.array(aux_src_params[:, 0:42], mask=1 - np.repeat(aux_src_mask, 42, axis=1))
masked_params = np.ma.array(
aux_src_params[:, 0:42],
mask=1 - np.repeat(aux_src_mask, 42, axis=1)
)
# Compute maximum and minimum values
spk_max[src_index, :] = np.maximum(spk_max[src_index, :], np.ma.max(masked_params, axis=0))
spk_min[src_index, :] = np.minimum(spk_min[src_index, :], np.ma.min(masked_params, axis=0))
spk_max[src_index, :] = np.maximum(
spk_max[src_index, :], np.ma.max(masked_params, axis=0)
)
spk_min[src_index, :] = np.minimum(
spk_min[src_index, :], np.ma.min(masked_params, axis=0)
)

# Append sequence params and masks to main datatables and masks
src_datatable.append(aux_src_params)
src_masks.append(aux_src_mask)
trg_datatable.append(aux_trg_params)
src_masks.append(aux_src_mask)
trg_masks.append(aux_trg_mask)

return (np.array(src_datatable),
Expand Down Expand Up @@ -308,10 +313,15 @@ def seq2seq_save_datatable(data_dir, datatable_out_file):

# Check that the data_dir ends in a '/'
try:
assert data_dir[len(data_dir) - 1] == '/'
assert data_dir[-1] == '/'
except AssertionError:
print("Please, make sure the data directory string ends with a '/'")

try:
assert type(datatable_out_file) == str
except AssertionError:
print("Please, make sure the data output filename is a string")

# Construct datatables and masks
(source_datatable,
source_masks,
Expand All @@ -326,11 +336,11 @@ def seq2seq_save_datatable(data_dir, datatable_out_file):
'seq2seq_basenames.list'
)

# Save dataset names and dataset arrays for elegant iteration
# Save dataset names and dataset arrays for elegant iteration when saving
data_dict = {
'src_datatable': source_datatable,
'src_mask': source_masks,
'trg_datatable': target_datatable,
'src_mask': source_masks,
'trg_mask': target_masks
}

Expand Down Expand Up @@ -378,8 +388,9 @@ def seq2seq2_load_datatable(datatable_file):
with h5py.File(datatable_file, 'r') as file:
# Load datasets
source_datatable = file['src_datatable'][:, :]
source_masks = file['src_mask'][:, :]
target_datatable = file['trg_datatable'][:, :]

source_masks = file['src_mask'][:, :]
target_masks = file['trg_mask'][:, :]

# Load max_seq_length attribute
Expand Down

0 comments on commit a086125

Please sign in to comment.