def create_dataset(speechfolder,
peaksfolder,
window,
stride,
file_slice=slice(0, 10)):
speechfiles = sorted(glob(os.path.join(speechfolder, '*.npy')))[file_slice]
peakfiles = sorted(glob(os.path.join(peaksfolder, '*.npy')))[file_slice]
speech_data = [np.load(f) for f in speechfiles]
peak_data = [np.load(f) for f in peakfiles]
speech_data = np.concatenate(speech_data)
peak_data = np.concatenate(peak_data)
indices = np.arange(len(speech_data))
speech_windowed_data = strided_app(speech_data, window, stride)
peak_windowed_data = strided_app(peak_data, window, stride)
indices = strided_app(indices, window, stride)
peak_distance = np.array([
np.nonzero(t)[0][0] if len(np.nonzero(t)[0]) != 0 else -1
for t in peak_windowed_data
])
peak_indicator = (peak_distance != -1) * 1.0
return speech_windowed_data, peak_distance, peak_indicator, indices, peak_data
def custom_loader(speechfolder, eggfolder, window, stride, select=None):
speechfiles = sorted(glob(os.path.join(speechfolder, "*.npy")))
eggfiles = sorted(glob(os.path.join(eggfolder, "*.npy")))
if select is not None:
ind = np.random.permutation(len(speechfiles))
ind = ind[:select]
speechfiles = [speechfiles[i] for i in ind]
eggfiles = [eggfiles[i] for i in ind]
print("Selected {} files".format(select))
speech_data = [np.load(f) for f in speechfiles]
egg_data = [np.load(f) for f in eggfiles]
for i in range(len(egg_data)):
egg_data[i] = egg_data[i] / np.max(np.abs(egg_data[i]))
for i in range(len(speech_data)):
speech_data[i] = speech_data[i] / np.max(np.abs(speech_data[i]))
speech_data = np.concatenate(speech_data)
egg_data = np.concatenate(egg_data)
speech_windowed_data = strided_app(speech_data, window, stride)
egg_windowed_data = strided_app(egg_data, window, stride)
return speech_windowed_data, egg_windowed_data
def _create_dataset(speechfiles, peakfiles, window, stride):
speech_data = [np.load(f) for f in speechfiles]
peak_data = [np.load(f) for f in peakfiles]
speech_data = np.concatenate(speech_data)
peak_data = np.concatenate(peak_data)
speech_windowed_data = strided_app(speech_data, window, stride)
peak_windowed_data = strided_app(peak_data, window, stride)
peak_distance = np.array([
np.nonzero(t)[0][0] if len(np.nonzero(t)[0]) != 0 else -1
for t in peak_windowed_data
])
peak_indicator = (peak_distance != -1) * 1.0
return speech_windowed_data, peak_distance, peak_indicator
def create_dataset(speechfolder,
peaksfolder,
window,
stride,
file_slice=slice(0, 10)):
speechfiles = sorted(glob(os.path.join(speechfolder, '*.npy')))[file_slice]
peakfiles = sorted(glob(os.path.join(peaksfolder, '*.npy')))[file_slice]
# speechfiles = speechfiles[:10]
# peakfiles = peakfiles[:10]
speech_data = [np.load(f) for f in speechfiles]
peak_data = [np.load(f) for f in peakfiles]
speech_data = np.concatenate(speech_data)
peak_data = np.concatenate(peak_data)
indices = np.arange(len(speech_data))
speech_windowed_data = strided_app(speech_data, window, stride)
peak_windowed_data = strided_app(peak_data, window, stride)
indices = strided_app(indices, window, stride)
return speech_windowed_data, indices, peak_data
def update_lists(self):
self.example_list = []
for i, clss in enumerate(self.class2file):
clss_file_list = np.random.permutation(self.class2file[clss])
idxs = strided_app(np.arange(len(clss_file_list)), 4, 4)
for idxs_list in idxs:
if len(idxs_list) == 4:
self.example_list.append(
[clss_file_list[file_idx] for file_idx in idxs_list])
self.example_list[-1].append(clss)
self.example_list[-1].append(self.clss2label[clss])