# Prepare sampler
frame_lengths = Mel.file_data_source.frame_lengths
sampler = PartialyRandomizedSimilarTimeLengthSampler(
frame_lengths, batch_size=hparams.batch_size)
# Dataset and Dataloader setup
dataset = PyTorchDataset(X, Mel, Y)
data_loader_dv3 = data_utils.DataLoader(
dataset, batch_size=hparams.batch_size,
num_workers=hparams.num_workers, sampler=sampler,
collate_fn=collate_fn, pin_memory=hparams.pin_memory)
print("dataloader for dv3 prepared")
dv3.train._frontend = getattr(frontend, hparams.frontend)
dv3_model = build_deepvoice_3(dv3_preset , checkpoint_dv3)
print("Built dv3!")
if use_cuda:
dv3_model = dv3_model.cuda()
dv3_optimizer = optim.Adam((dv3_model.get_trainable_parameters(),
lr=hparams.initial_learning_rate, betas=(
hparams.adam_beta1, hparams.adam_beta2),
eps=hparams.adam_eps, weight_decay=hparams.weight_decay)
log_event_path = "log/run-test" + str(datetime.now()).replace(" ", "_")
print("Log event path for dv3: {}".format(log_event_path))
writer_dv3 = SummaryWriter(log_dir=log_event_path)
# ENCODER
#if i%8000=0:
#scheduler.step()
def download_file(file_name=None):
from google.colab import files
files.download(file_name)
batch_size = 64
if __name__ == "__main__":
#Load Deep Voice 3
# Pre Trained Model
dv3_model = build_deepvoice_3(True)
all_speakers = get_cloned_voices()
print("Cloning Texts are produced")
speaker_embed = get_speaker_embeddings(dv3_model)
#
encoder = build_encoder()
print("Encoder is built!")
speech_data = Speech_Dataset(all_speakers, speaker_embed)
#for i in range(5):
# sample = speech_data[i]
# print(sample[0].shape, sample[1].shape)
# print(sample[0], sample[1])