def _single_conversion(filename, model, one_hot_emo):
'''
THIS WON'T WORK RIGHT NOW, USE THE WORLD CONVERSION LOOP IN MAIN
Call only from __main__ section in this module. Generates sample converted
into each emotion.
(str) filename - name.wav file to be converted
(StarGAN-emo-VC1) model - pretrained model to perform conversion
(torch.Tensor(long)) one_hot_emo - one hot encoding of emotion to convert to
'''
wav, labels = pp.get_wav_and_labels(filenames[5],
config['data']['dataset_dir'])
wav = np.array(wav, dtype=np.double)
f0, ap, sp, coded_sp = preprocess_world.cal_mcep(wav)
coded_sp = coded_sp.T
coded_sp_torch = torch.Tensor(coded_sp).unsqueeze(0).unsqueeze(0).to(
device=device)
fake = model.G(coded_sp_torch, one_hot_emo.unsqueeze(0))
fake = fake.squeeze()
print("Sampled size = ", fake.size())
converted_sp = fake.cpu().detach().numpy()
converted_sp = np.array(converted_sp, dtype=np.float64)
sample_length = converted_sp.shape[0]
if sample_length != ap.shape[0]:
ap = np.ascontiguousarray(ap[0:sample_length, :], dtype=np.float64)
f0 = np.ascontiguousarray(f0[0:sample_length], dtype=np.float64)
f0 = np.ascontiguousarray(f0[20:-20], dtype=np.float64)
ap = np.ascontiguousarray(ap[20:-20, :], dtype=np.float64)
converted_sp = np.ascontiguousarray(converted_sp[40:-40, :],
dtype=np.float64)
coded_sp = np.ascontiguousarray(coded_sp[20:-20, :], dtype=np.float64)
target = np.argmax(one_hot_emo)
out_name = filename[:-4] + str(labels[1]) + "to" + target + ".wav"
audio_utils.save_world_wav([f0, ap, sp, converted_sp],
model.name + '_converted', out_name)
def sample_world(self):
'''
Passes each performance sample through G for every target emotion. They
are saved to 'config(sample_dir)/model_name/filename-<emo>to<trg>.png + .npy'
'''
print("Saving world samples...")
self.model.to_device(device=self.device)
self.model.set_eval_mode()
# Make one-hot vector for each emotion category
num_emos = self.config['model']['num_classes']
emo_labels = torch.Tensor(range(0, num_emos)).long()
emo_targets = F.one_hot(
emo_labels, num_classes=num_emos).float().to(device=self.device)
for tag, val in self.sample_set.get_set().items():
# tag is filename, val is [mel, labels, spec]
f0_real = np.copy(val[0])
ap_real = np.copy(val[1])
sp = np.copy(val[2])
coded_sp = torch.Tensor.clone(val[3])
labels = torch.Tensor.clone(val[4])
coded_sp = val[3].unsqueeze(0).unsqueeze(0).to(device=self.device)
with torch.no_grad():
# print(emo_targets)
for i in range(0, emo_targets.size(0)):
f0 = np.copy(f0_real)
ap = np.copy(ap_real)
fake = self.model.G(coded_sp, emo_targets[i].unsqueeze(0))
filename_wav = tag[0:-4] + "_" + str(int(labels[0].item())) + "to" + \
str(emo_labels[i].item()) + '_i=' +\
str(self.current_iter) + ".wav"
fake = fake.squeeze()
print("Sampled size = ", fake.size())
# f = fake.data()
converted_sp = fake.cpu().numpy()
converted_sp = np.array(converted_sp, dtype=np.float64)
sample_length = converted_sp.shape[0]
if sample_length != ap.shape[0]:
ap = np.ascontiguousarray(ap[0:sample_length, :],
dtype=np.float64)
f0 = np.ascontiguousarray(f0[0:sample_length],
dtype=np.float64)
f0 = np.ascontiguousarray(f0[40:-40], dtype=np.float64)
ap = np.ascontiguousarray(ap[40:-40, :], dtype=np.float64)
converted_sp = np.ascontiguousarray(
converted_sp[40:-40, :], dtype=np.float64)
# print("ap shape = ", val[1].shape)
# print("f0 shape = ", val[0].shape)
audio_utils.save_world_wav([f0, ap, sp, converted_sp],
self.model_name, filename_wav)
sample_length = converted_sp.shape[0]
if sample_length != ap.shape[0]:
# coded_sp_temp_copy = np.ascontiguousarray(coded_sp_temp_copy[0:sample_length, :], dtype = np.float64)
ap = np.ascontiguousarray(ap[0:sample_length, :], dtype = np.float64)
f0 = np.ascontiguousarray(f0[0:sample_length], dtype = np.float64)
f0 = np.ascontiguousarray(f0[20:-20], dtype = np.float64)
ap = np.ascontiguousarray(ap[20:-20,:], dtype = np.float64)
converted_sp = np.ascontiguousarray(converted_sp[20:-20,:], dtype = np.float64)
# coded_sp_temp_copy = np.ascontiguousarray(coded_sp_temp_copy[40:-40,:], dtype = np.float64)
# print("ap shape = ", ap.shape)
# print("f0 shape = ", f0.shape)
print(converted_sp.shape)
it = str(args.iteration)[0:3]
audio_utils.save_world_wav([f0,ap,sp,converted_sp], args.out_dir +"_"+ it, filename_wav)
# print(f, " converted.")
if (file_num+1) % 20 == 0:
print(file_num+1, " done.")
########################################
# MEL CONVERSION LOOP #
########################################
### NEVER IMPLEMENTED AS ENDED UP NOT USING MEL SPECTROGRAMS
# Make .npy arrays
# Make audio
# Make spec plots
# Save all to directory