def generate_decoders_data(data_dir, data_extension):
# File reader won't work, so I need to load audio files into external_source manually
fnames = test_utils.filter_files(data_dir, data_extension)
nfiles = len(fnames)
_input_epoch = [
list(
map(lambda fname: test_utils.read_file_bin(fname),
fnames[:nfiles // 3])),
list(
map(lambda fname: test_utils.read_file_bin(fname),
fnames[nfiles // 3:nfiles // 2])),
list(
map(lambda fname: test_utils.read_file_bin(fname),
fnames[nfiles // 2:])),
]
# Since we pack buffers into ndarray, we need to pad samples with 0.
input_epoch = []
for inp in _input_epoch:
max_len = max(sample.shape[0] for sample in inp)
inp = map(
lambda sample: np.pad(sample, (0, max_len - sample.shape[0])), inp)
input_epoch.append(np.stack(list(inp)))
input_epoch = list(
map(lambda batch: np.reshape(batch, batch.shape), input_epoch))
return input_epoch
def generate_decoders_data(data_dir, data_extension):
# File reader won't work, so I need to load audio files into external_source manually
fnames = test_utils.filter_files(data_dir, data_extension)
nfiles = len(fnames)
# TODO(janton): Workaround for audio data (not enough samples)
# To be removed when more audio samples are added
for i in range(len(fnames), 10): # At least 10 elements
fnames.append(fnames[-1])
nfiles = len(fnames)
_input_epoch = [
list(map(lambda fname: test_utils.read_file_bin(fname), fnames[:nfiles // 3])),
list(map(lambda fname: test_utils.read_file_bin(fname),
fnames[nfiles // 3: nfiles // 2])),
list(map(lambda fname: test_utils.read_file_bin(fname), fnames[nfiles // 2:])),
]
# Since we pack buffers into ndarray, we need to pad samples with 0.
input_epoch = []
for inp in _input_epoch:
max_len = max(sample.shape[0] for sample in inp)
inp = map(lambda sample: np.pad(sample, (0, max_len - sample.shape[0])), inp)
input_epoch.append(np.stack(list(inp)))
input_epoch = list(map(lambda batch: np.reshape(batch, batch.shape), input_epoch))
return input_epoch