def __init__(self, name):
self._action_name = name
self.tacotron2 = Tacotron2()
self.feedback = tacotron2_ros.msg.TTSFeedback()
self.result = tacotron2_ros.msg.TTSResult()
self.server = actionlib.SimpleActionServer(self._action_name, tacotron2_ros.msg.TTSAction, execute_cb=self.execute_cb, auto_start=False)
self.server.start()
def evluate(test_data_generator, vocab_size, config, test_batchs):
# 模型初始化
tacotron2 = Tacotron2(vocab_size, config)
path = config.checkpoingt_dir
checkpoint = load_checkpoint(tacotron2, path, config)
print('已恢复至最新的检查点！')
j = 0
score_sum = 0
for batch, (input_ids, mel_gts,
mel_len_wav) in zip(range(test_batchs), test_data_generator):
for i in range(input_ids.shape[0]):
new_input_ids = input_ids[i]
new_input_ids = tf.expand_dims(new_input_ids, axis=0)
mel_outputs, mel_outputs_postnet, gate_outputs, alignments = tacotron2.inference(
new_input_ids)
mel2 = mel_gts[i]
mel2 = tf.expand_dims(mel2, axis=0)
mel2 = tf.transpose(mel2, [0, 2, 1])
score = spec_distance(mel_outputs_postnet, mel2)
score_sum += score
j = j + 1
print('第{}个样本的欧式距离为：{}'.format(j, score))
print("样本平均欧式距离为：", score_sum / j)
return data
if __name__ == "__main__":
config = Tacotron2Config()
# 检查点路径
path = config.checkpoingt_dir
# 字典路径
save_path_dictionary = config.save_path_dictionary
# 恢复字典
tokenizer, vocab_size = get_tokenizer_keras(save_path_dictionary)
# 模型初始化
tacotron2 = Tacotron2(vocab_size, config)
# 加载检查点
checkpoint = load_checkpoint(tacotron2, path, config)
print('已恢复至最新的检查点！')
# 用于命名
i = 0
# 抓取文本数据
while True:
i = i+1
b = str(i)
print("请输入您要合成的话，输入ESC结束：")
seq = input()
if seq == 'ESC':
break
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch
import numpy as np
from tacotron2 import Tacotron2
from .args import DefaultArgument
batch_size = 3
seq_length = 3
inputs = torch.LongTensor(
np.arange(batch_size * seq_length).reshape(batch_size, seq_length))
input_lengths = torch.LongTensor([3, 3, 2])
targets = torch.FloatTensor(batch_size, 100, 80).uniform_(-0.1, 0.1)
args = DefaultArgument()
model = Tacotron2(args)
output = model(inputs, targets, input_lengths)
print(model)
print(output)
args = parser.parse_args()
if args.wandb:
import wandb
wandb.init(project='tacotron 2')
import warnings
warnings.simplefilter(action='ignore', category=FutureWarning)
torch.backends.cudnn.deterministic = True
device = 'cuda'
model = Tacotron2(conf)
dataset = TTSDataset('kss.lmdb')
indices = list(range(len(dataset)))
random.seed(17)
random.shuffle(indices)
train_ind = indices[200:]
valid_ind = indices[:200]
train_set = Subset(dataset, train_ind)
valid_set = Subset(dataset, valid_ind)
train_loader = DataLoader(train_set,
batch_size=64,
shuffle=True,
num_workers=4,
state, losses = optimize_step(sub_inputs, state, sub_target)
state = to_np(state)
for name, value in losses.items():
total_loss.setdefault(name, 0.)
total_loss[name] += value * tf.cast((end - start), tf.float32)
total_loss = {
k: v / tf.cast(n_frames, tf.float32)
for k, v in total_loss.items()
}
return total_loss
model = Tacotron2(vocab_size=148)
loss_fn = TacotronLoss()
optimizer = tf.keras.optimizers.Adam()
batch_size = 64
length = 400
txt_inp = tf.ones((batch_size, 150), dtype=tf.int32)
mel_inp = tf.ones((batch_size, length, 80), dtype=tf.float32)
gate_inp = tf.ones((batch_size, length), dtype=tf.float32)
_ = model((txt_inp, mel_inp))
batch = ((txt_inp, mel_inp), (mel_inp, gate_inp))
loss = train_step(model, loss_fn, optimizer, batch)
def __init__(self):
rospy.init_node("tacotron2")
self.tacotron2 = Tacotron2()
sub = rospy.Subscriber("/tacotron2/tts", String, self.callback)
rospy.spin()