# model.set_r(hp.tts_r)
optimiser = optim.Adam(model.parameters())
current_step = model.get_step()
if not force_gta:
for session in hp.tts_schedule:
r, lr, max_step, batch_size = session
if current_step < max_step:
train_set, attn_example = get_tts_dataset(
paths.data, batch_size, r)
model.set_r(r)
training_steps = max_step - current_step
simple_table([(f'Steps with r={r}',
str(training_steps // 1000) + 'k Steps'),
('Batch Size', batch_size),
('Learning Rate', lr),
('Outputs/Step (r)', model.get_r())])
tts_train_loop(model, optimiser, train_set, lr, training_steps,
attn_example)
print('Training Complete.')
n_mels=hp.num_mels,
fft_bins=hp.num_mels,
postnet_dims=hp.tts_postnet_dims,
encoder_K=hp.tts_encoder_K,
lstm_dims=hp.tts_lstm_dims,
postnet_K=hp.tts_postnet_K,
num_highways=hp.tts_num_highways,
dropout=hp.tts_dropout).cuda()
paths = Paths(hp.data_path, hp.voc_model_id, hp.tts_model_id)
model.restore(paths.tts_latest_weights)
optimiser = optim.Adam(model.parameters())
train_set = get_tts_dataset(paths.data, batch_size)
if not force_gta:
total_steps = 10_000_000 if force_train else hp.tts_total_steps
simple_table([
('Remaining', str(
(total_steps - model.get_step()) // 1000) + 'k Steps'),
('Batch Size', batch_size), ('Learning Rate', lr)
])
tts_train_loop(model, optimiser, train_set, lr, total_steps)
print('Training Complete.')
print(