class Solver(object):
"""训练操作的说明"""
# 参数为数据加载器与配置
def __init__(self, data_loader, config):
# 进行赋值
self.config = config
self.data_loader = data_loader
# 模型配置
# 赋值三个损失函数，循环损失，域分类损失，身份映射损失的衡量参数
self.lambda_cycle = config.lambda_cycle
self.lambda_cls = config.lambda_cls
self.lambda_identity = config.lambda_identity
# 训练配置
# 数据文件路由
self.data_dir = config.data_dir
# 测试文件路由
self.test_dir = config.test_dir
# 批处理大小
self.batch_size = config.batch_size
# 训练判别器D的总迭代次数
self.num_iters = config.num_iters
# 衰减学习率的迭代次数
self.num_iters_decay = config.num_iters_decay
# 生成器G的学习频率
self.g_lr = config.g_lr
# 判别器D的学习频率
self.d_lr = config.d_lr
# 域分类器C的学习频率
self.c_lr = config.c_lr
# 每次G更新时的D更新次数
self.n_critic = config.n_critic
# Adam优化器的beta1，2参数
self.beta1 = config.beta1
self.beta2 = config.beta2
# 从此步骤恢复培训
self.resume_iters = config.resume_iters
# 测试配置
# 从这个步骤开始训练模型
self.test_iters = config.test_iters
# ast.literal_eval为解析函数，并安全地进行类型转换
# 目标发音者
self.trg_speaker = ast.literal_eval(config.trg_speaker)
# 源发音者
self.src_speaker = config.src_speaker
# 其他配置
# 是否使用tensorboard记录
self.use_tensorboard = config.use_tensorboard
# torch.device代表将torch.Tensor分配到的设备的对象。torch.device包含一个设备类型（‘cpu’或‘cuda’）和可选的设备序号。
# 是使用cuda还是cpu计算
self.device = torch.device(
'cuda:0' if torch.cuda.is_available() else 'cpu')
# 将speakers标签列表二值化
self.spk_enc = LabelBinarizer().fit(speakers)
# 字典
# 记录目录
self.log_dir = config.log_dir
# 样本目录
self.sample_dir = config.sample_dir
# 模型目录
self.model_save_dir = config.model_save_dir
# 输出目录
self.result_dir = config.result_dir
# 步长
# 记录步长
self.log_step = config.log_step
# 采样步长
self.sample_step = config.sample_step
# 模型保存间隔步长
self.model_save_step = config.model_save_step
# 学习率更新步长
self.lr_update_step = config.lr_update_step
# 建立模型与tensorboard.
self.build_model()
if self.use_tensorboard:
# 使用tensorboard记录器
self.build_tensorboard()
# 赋值三个模型器
def build_model(self):
# 将模型赋值给类属性
self.G = Generator()
self.D = Discriminator()
self.C = DomainClassifier()
# torch.optim.Adam用于实现Adam算法
# Adam(Adaptive Moment Estimation)本质上是带有动量项的RMSprop，它利用梯度的一阶矩估计和二阶矩估计动态调整每个参数的学习率。
# 它的优点主要在于经过偏置校正后，每一次迭代学习率都有个确定范围，使得参数比较平稳。
# params (iterable) – 待优化参数的iterable或者是定义了参数组的dict
# lr (float, 可选) – 学习率（默认：1e-3）
# 同样也称为学习率或步长因子，它控制了权重的更新比率（如 0.001）。较大的值（如 0.3）在学习率更新前会有更快的初始学习，而较小的值（如 1.0E-5）会令训练收敛到更好的性能。
# betas (Tuple[float, float], 可选) – 用于计算梯度以及梯度平方的运行平均值的系数（默认：0.9，0.999）betas = （beta1，beta2）
# beta1：一阶矩估计的指数衰减率（如 0.9）。
# beta2：二阶矩估计的指数衰减率（如 0.999）。该超参数在稀疏梯度（如在 NLP 或计算机视觉任务中）中应该设置为接近 1 的数。
# eps (float, 可选) – 为了增加数值计算的稳定性而加到分母里的项（默认：1e-8）epsilon：该参数是非常小的数，其为了防止在实现中除以零（如 10E-8）。
# weight_decay (float, 可选) – 权重衰减（L2级惩罚）（默认: 0）
self.g_optimizer = torch.optim.Adam(self.G.parameters(), self.g_lr,
[self.beta1, self.beta2])
self.d_optimizer = torch.optim.Adam(self.D.parameters(), self.d_lr,
[self.beta1, self.beta2])
self.c_optimizer = torch.optim.Adam(self.C.parameters(), self.c_lr,
[self.beta1, self.beta2])
self.print_network(self.G, 'G')
self.print_network(self.D, 'D')
self.print_network(self.C, 'C')
# Module.to方法用来移动和转换参数与缓冲区，类似于torch.Tensor.to，但是仅支持float类型
# 这里将模型转移到GPU运算
self.G.to(self.device)
self.D.to(self.device)
self.C.to(self.device)
def print_network(self, model, name):
"""打印出网络的相关信息"""
num_params = 0
# Module.parameters()获取网络的参数
# 计算模型网络对应的参数频次
for p in model.parameters():
# numel返回数组中元素的个数
num_params += p.numel()
print(model)
print(name)
print("参数的个数为：{}".format(num_params))
# 使用tensorboard
def build_tensorboard(self):
"""建立一个tensorboard记录器"""
from logger import Logger
# 建立一个记录器，传入记录地址
self.logger = Logger(self.log_dir)
def update_lr(self, g_lr, d_lr, c_lr):
"""生成器、判别器和域分类器的衰减学习率"""
for param_group in self.g_optimizer.param_groups:
param_group['lr'] = g_lr
for param_group in self.d_optimizer.param_groups:
param_group['lr'] = d_lr
for param_group in self.c_optimizer.param_groups:
param_group['lr'] = c_lr
# 训练方法
def train(self):
# 衰减的学习率缓存
g_lr = self.g_lr
d_lr = self.d_lr
c_lr = self.c_lr
# 开始训练步骤数为0
start_iters = 0
# 如果存在就跳过
if self.resume_iters:
pass
# 调用定义的个性化标准化方法
norm = Normalizer()
# iter用来生成迭代器，这里用来迭代加载数据集
data_iter = iter(self.data_loader)
print('开始训练......')
# 记录当前时间，now函数取当前时间
start_time = datetime.now()
# 利用总迭代次数来进行遍历
for i in range(start_iters, self.num_iters):
# =================================================================================== #
# 1.预处理输入数据 #
# =================================================================================== #
# 获取真实的图像和对应标签标签
try:
# next方法为迭代下一个迭代器
# 利用自定义的加载器获取真实x值，发音者标签在组中索引与源标签
x_real, speaker_idx_org, label_org = next(data_iter)
except:
# 如果迭代器有问题就再转换为迭代器一次然后迭代
data_iter = iter(self.data_loader)
x_real, speaker_idx_org, label_org = next(data_iter)
# 随机生成目标域标签
# torch.randperm返回一个从0到参数-1范围的随机数组
# 因为标签二值化了，所以这里的标签是10组成的，所以一共有label_org.size(0)个标签
# 获得的是随机索引
rand_idx = torch.randperm(label_org.size(0))
# 根据随机数作为源标签的索引作为目标标签数
label_trg = label_org[rand_idx]
# 同理得到随机目标发音者
speaker_idx_trg = speaker_idx_org[rand_idx]
# to表示使用cpu或者gpu运行
x_real = x_real.to(self.device) # 输入数据
label_org = label_org.to(self.device) # 源域one-hot格式标签
label_trg = label_trg.to(self.device) # 目标域ont-hot格式标签
speaker_idx_org = speaker_idx_org.to(self.device) # 源域标签
speaker_idx_trg = speaker_idx_trg.to(self.device) # 目标域标签
# =================================================================================== #
# 2.训练判别器 #
# =================================================================================== #
# 用真实音频数据计算损失
# nn.CrossEntropyLoss()为交叉熵损失函数，但是不是普通的形式，而是主要是将softmax-log-NLLLoss合并到一块得到的结果。
CELoss = nn.CrossEntropyLoss()
# 调用分类器计算真实数据
cls_real = self.C(x_real)
# 计算对应的域分类损失，即用交叉熵实现
cls_loss_real = CELoss(input=cls_real, target=speaker_idx_org)
# 重置缓冲区，具体实现在下面
self.reset_grad()
# tensor.backward为自动求导函数
cls_loss_real.backward()
# optimizer.step这个方法会更新模型所有的参数以提升学习率，一般在backward函数后根据其计算的梯度来更新参数
self.c_optimizer.step()
# 记录中
loss = {}
# 从真实域分类损失张量中获取元素值
# item()得到一个元素张量里面的元素值
loss['C/C_loss'] = cls_loss_real.item()
# 基于源数据的D判断结果
out_r = self.D(x_real, label_org)
# 用假音频帧计算损失
# 根据真实样本与目标标签生成生成样本
x_fake = self.G(x_real, label_trg)
# detach截断反向传播的梯度流，从而让梯度不影响判别器D
# 基于生成样本的D判断结果
out_f = self.D(x_fake.detach(), label_trg)
# torch.nn.Function.binary_cross_entropy_with_logits度量目标逻辑和输出逻辑之间的二进制交叉熵的函数
# 接受任意形状的输入，target要求与输入形状一致。切记：target的值必须在[0,N-1]之间，其中N为类别数，否则会出现莫名其妙的错误，比如loss为负数。
# 计算其实就是交叉熵，不过输入不要求在0，1之间，该函数会自动添加sigmoid运算
# 返回一个填充了标量值1的张量，其大小与输入相同。torch.ones_like(input)
# 相当于torch.ones(input.size(), dtype=input.dtype, layout=input.layout, device=input.device)
# binary_cross_entropy_with_logits和binary_cross_entropy的区别
# 有一个（类）损失函数名字中带了with_logits. 而这里的logits指的是,该损失函数已经内部自带了计算logit的操作，
# 无需在传入给这个loss函数之前手动使用sigmoid/softmax将之前网络的输入映射到[0,1]之间
d_loss_t = F.binary_cross_entropy_with_logits(input=out_f,target=torch.zeros_like(out_f, dtype=torch.float)) + \
F.binary_cross_entropy_with_logits(input=out_r, target=torch.ones_like(out_r, dtype=torch.float))
# 生成样本的分类结果
out_cls = self.C(x_fake)
# 交叉熵计算生成样本的域分类损失
d_loss_cls = CELoss(input=out_cls, target=speaker_idx_trg)
# 计算梯度惩罚的损失
alpha = torch.rand(x_real.size(0), 1, 1, 1).to(self.device)
# 计算x_hat
# requires_grad_设置积分方法，将requires_grad是否积分的属性设置为真
# 取一个随机数混合真实样本和生成样本得到一个x尖
x_hat = (alpha * x_real.data +
(1 - alpha) * x_fake.data).requires_grad_(True)
# 计算混合样本和目标标签的判别结果
out_src = self.D(x_hat, label_trg)
# 调用自定义方法得到处理导数后的数据
d_loss_gp = self.gradient_penalty(out_src, x_hat)
# 计算判别器的总体损失
d_loss = d_loss_t + self.lambda_cls * d_loss_cls + 5 * d_loss_gp
# 调用自定义方法重置梯度变化缓冲区
self.reset_grad()
# 对D的损失求导
d_loss.backward()
# 更新模型判别器D参数
self.d_optimizer.step()
# loss['D/d_loss_t'] = d_loss_t.item()
# loss['D/loss_cls'] = d_loss_cls.item()
# loss['D/D_gp'] = d_loss_gp.item()
# 获取判别器损失
loss['D/D_loss'] = d_loss.item()
# =================================================================================== #
# 3.训练生成器 #
# =================================================================================== #
# 进行模运算，判读更新时间
if (i + 1) % self.n_critic == 0:
# 源至目标域
# 利用真实样本和目标标签生成生成样本
x_fake = self.G(x_real, label_trg)
# 判别生成样本与目标标签
g_out_src = self.D(x_fake, label_trg)
# 将生成与目标标签的损失与相同大小纯1张量计算交叉熵得到生成G损失
g_loss_fake = F.binary_cross_entropy_with_logits(
input=g_out_src,
target=torch.ones_like(g_out_src, dtype=torch.float))
# 得到真实样本通过域分类器得到的类别
out_cls = self.C(x_real)
# 计算C计算类别与输入的类别的交叉熵损失即G的分类损失
g_loss_cls = CELoss(input=out_cls, target=speaker_idx_org)
# 目标至源域
# 通过G将生成样本转换为源标签
x_reconst = self.G(x_fake, label_org)
# 得到循环一致性损失，即通过G转回来的损失，按道理这两个是同样的
# l1_loss为L1损失函数，即平均绝对误差
g_loss_rec = F.l1_loss(x_reconst, x_real)
# 源到源域(身份一致性损失).
# 通过真实样本与源标签生成，按道理也是生成x_real
x_fake_iden = self.G(x_real, label_org)
# 利用L1损失函数计算
id_loss = F.l1_loss(x_fake_iden, x_real)
# 后退和优化
# 得到生成器的总体损失函数
g_loss = g_loss_fake + self.lambda_cycle * g_loss_rec +\
self.lambda_cls * g_loss_cls + self.lambda_identity * id_loss
# 重置梯度变化缓冲区
self.reset_grad()
# 对G损失求导
g_loss.backward()
# 更新生成器参数
self.g_optimizer.step()
# 记录对应的损失
loss['G/loss_fake'] = g_loss_fake.item()
loss['G/loss_rec'] = g_loss_rec.item()
loss['G/loss_cls'] = g_loss_cls.item()
loss['G/loss_id'] = id_loss.item()
loss['G/g_loss'] = g_loss.item()
# =================================================================================== #
# 4.其他 #
# =================================================================================== #
# 打印训练相关信息
if (i + 1) % self.log_step == 0:
# 得到训练时间
et = datetime.now() - start_time
# 截取后面的时间段
et = str(et)[:-7]
# 耗时与迭代次数
log = "耗时：[{}], 迭代次数：[{}/{}]".format(et, i + 1, self.num_iters)
# 打印对应损失值
for tag, value in loss.items():
log += ", {}: {:.4f}".format(tag, value)
print(log)
# 如果调用tensorboard来记录训练过程
if self.use_tensorboard:
for tag, value in loss.items():
# 添加到log中
self.logger.scalar_summary(tag, value, i + 1)
# 翻译固定数据进行调试
if (i + 1) % self.sample_step == 0:
# torch.no_grad是一个上下文管理器，被该语句包括起来的部分将不会track 梯度
# 所有依赖他的tensor会全部变成True，反向传播时就不会自动求导了，反向传播就不会保存梯度，因此大大节约了显存或者说内存。
with torch.no_grad():
# 调用自定义方法，定义一个路由，并随机选取一个发音者作为测试数据
d, speaker = TestSet(self.test_dir).test_data()
# random.choice返回参数的随机项
# 随机在speakers中选择一个不是目标的发音者
target = random.choice(
[x for x in speakers if x != speaker])
# 将二值化的标签组取出第一个作为目标
# LabelBinary.transfrom方法将复杂类标签转换为二进制标签
label_t = self.spk_enc.transform([target])[0]
# np.asarray将python原生列表或元组形式的现有数据来创建numpy数组
label_t = np.asarray([label_t])
# 取出字典中的文件名与内容
for filename, content in d.items():
f0 = content['f0']
ap = content['ap']
# 调用自定义方法处理对应的数据
sp_norm_pad = self.pad_coded_sp(
content['coded_sp_norm'])
convert_result = []
for start_idx in range(
0, sp_norm_pad.shape[1] - FRAMES + 1, FRAMES):
one_seg = sp_norm_pad[:,
start_idx:start_idx + FRAMES]
one_seg = torch.FloatTensor(one_seg).to(
self.device)
one_seg = one_seg.view(1, 1, one_seg.size(0),
one_seg.size(1))
l = torch.FloatTensor(label_t)
one_seg = one_seg.to(self.device)
l = l.to(self.device)
one_set_return = self.G(one_seg,
l).data.cpu().numpy()
one_set_return = np.squeeze(one_set_return)
one_set_return = norm.backward_process(
one_set_return, target)
convert_result.append(one_set_return)
convert_con = np.concatenate(convert_result, axis=1)
convert_con = convert_con[:,
0:content['coded_sp_norm'].
shape[1]]
contigu = np.ascontiguousarray(convert_con.T,
dtype=np.float64)
decoded_sp = decode_spectral_envelope(contigu,
SAMPLE_RATE,
fft_size=FFTSIZE)
f0_converted = norm.pitch_conversion(
f0, speaker, target)
wav = synthesize(f0_converted, decoded_sp, ap,
SAMPLE_RATE)
name = f'{speaker}-{target}_iter{i+1}_{filename}'
path = os.path.join(self.sample_dir, name)
print(f'[save]:{path}')
librosa.output.write_wav(path, wav, SAMPLE_RATE)
# 保存模型检查点
if (i + 1) % self.model_save_step == 0:
G_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(i + 1))
D_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(i + 1))
C_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(i + 1))
torch.save(self.G.state_dict(), G_path)
torch.save(self.D.state_dict(), D_path)
torch.save(self.C.state_dict(), C_path)
print('Saved model checkpoints into {}...'.format(
self.model_save_dir))
# 衰减学习率
if (i + 1) % self.lr_update_step == 0 and (i + 1) > (
self.num_iters - self.num_iters_decay):
g_lr -= (self.g_lr / float(self.num_iters_decay))
d_lr -= (self.d_lr / float(self.num_iters_decay))
c_lr -= (self.c_lr / float(self.num_iters_decay))
self.update_lr(g_lr, d_lr, c_lr)
print('Decayed learning rates, g_lr: {}, d_lr: {}.'.format(
g_lr, d_lr))
def gradient_penalty(self, y, x):
"""计算梯度惩罚: (L2_norm(dy/dx) - 1)**2."""
# 根据标签的纬度创建出权重矩阵
weight = torch.ones(y.size()).to(self.device)
# torch.autograd.grad计算并返回outputs对inputs的梯度dy/dx
dydx = torch.autograd.grad(
outputs=y,
inputs=x,
# 雅可比向量积中的“向量”，用来将梯度向量转换为梯度标量，并可以衡量y梯度的各个数据的权重
grad_outputs=weight,
retain_graph=True,
# 保持计算的导数图
create_graph=True,
only_inputs=True)[0]
# 将导数变为二维的数据并保持行数
dydx = dydx.view(dydx.size(0), -1)
# 计算导数的L2范数
dydx_l2norm = torch.sqrt(torch.sum(dydx**2, dim=1))
return torch.mean((dydx_l2norm - 1)**2)
def reset_grad(self):
"""重置梯度变化缓冲区"""
self.g_optimizer.zero_grad()
self.d_optimizer.zero_grad()
self.c_optimizer.zero_grad()
def restore_model(self, resume_iters):
"""重置训练好的发生器和鉴别器"""
print('从{}步骤开始加载训练过的模型...'.format(resume_iters))
G_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(resume_iters))
D_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(resume_iters))
C_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(resume_iters))
self.G.load_state_dict(
torch.load(G_path, map_location=lambda storage, loc: storage))
self.D.load_state_dict(
torch.load(D_path, map_location=lambda storage, loc: storage))
self.C.load_state_dict(
torch.load(C_path, map_location=lambda storage, loc: storage))
@staticmethod
def pad_coded_sp(coded_sp_norm):
f_len = coded_sp_norm.shape[1]
if f_len >= FRAMES:
pad_length = FRAMES - (f_len - (f_len // FRAMES) * FRAMES)
elif f_len < FRAMES:
pad_length = FRAMES - f_len
sp_norm_pad = np.hstack(
(coded_sp_norm, np.zeros((coded_sp_norm.shape[0], pad_length))))
return sp_norm_pad
def test(self):
"""用StarGAN处理音频数据"""
# 加载训练生成器
self.restore_model(self.test_iters)
norm = Normalizer()
# 设置数据加载器
d, speaker = TestSet(self.test_dir).test_data(self.src_speaker)
targets = self.trg_speaker
for target in targets:
print(target)
assert target in speakers
label_t = self.spk_enc.transform([target])[0]
label_t = np.asarray([label_t])
with torch.no_grad():
for filename, content in d.items():
f0 = content['f0']
ap = content['ap']
sp_norm_pad = self.pad_coded_sp(content['coded_sp_norm'])
convert_result = []
for start_idx in range(0,
sp_norm_pad.shape[1] - FRAMES + 1,
FRAMES):
one_seg = sp_norm_pad[:, start_idx:start_idx + FRAMES]
one_seg = torch.FloatTensor(one_seg).to(self.device)
one_seg = one_seg.view(1, 1, one_seg.size(0),
one_seg.size(1))
l = torch.FloatTensor(label_t)
one_seg = one_seg.to(self.device)
l = l.to(self.device)
one_set_return = self.G(one_seg, l).data.cpu().numpy()
one_set_return = np.squeeze(one_set_return)
one_set_return = norm.backward_process(
one_set_return, target)
convert_result.append(one_set_return)
convert_con = np.concatenate(convert_result, axis=1)
convert_con = convert_con[:, 0:content['coded_sp_norm'].
shape[1]]
contigu = np.ascontiguousarray(convert_con.T,
dtype=np.float64)
decoded_sp = decode_spectral_envelope(contigu,
SAMPLE_RATE,
fft_size=FFTSIZE)
f0_converted = norm.pitch_conversion(f0, speaker, target)
wav = synthesize(f0_converted, decoded_sp, ap, SAMPLE_RATE)
name = f'{speaker}-{target}_iter{self.test_iters}_{filename}'
path = os.path.join(self.result_dir, name)
print(f'[保存]:{path}')
librosa.output.write_wav(path, wav, SAMPLE_RATE)
class Solver(object):
"""Solver for training and testing StarGAN."""
def __init__(self, train_loader, test_loader, config):
"""Initialize configurations."""
# Data loader.
self.train_loader = train_loader
self.test_loader = test_loader
self.sampling_rate = config.sampling_rate
# Model configurations.
self.num_speakers = config.num_speakers
self.lambda_cls = config.lambda_cls
self.lambda_rec = config.lambda_rec
self.lambda_gp = config.lambda_gp
self.lambda_id = config.lambda_id
# Training configurations.
self.batch_size = config.batch_size
self.num_iters = config.num_iters
self.num_iters_decay = config.num_iters_decay
self.g_lr = config.g_lr
self.d_lr = config.d_lr
self.c_lr = config.c_lr
self.n_critic = config.n_critic
self.beta1 = config.beta1
self.beta2 = config.beta2
self.resume_iters = config.resume_iters
# Test configurations.
self.test_iters = config.test_iters
# Miscellaneous.
self.use_tensorboard = config.use_tensorboard
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
# Directories.
self.log_dir = config.log_dir
self.sample_dir = config.sample_dir
self.model_save_dir = config.model_save_dir
# Step size.
self.log_step = config.log_step
self.sample_step = config.sample_step
self.model_save_step = config.model_save_step
self.lr_update_step = config.lr_update_step
# Build the model and tensorboard.
self.build_model()
if self.use_tensorboard:
self.build_tensorboard()
def build_model(self):
"""Create a generator and a discriminator."""
self.generator = Generator()
self.discriminator = Discriminator(num_speakers=self.num_speakers)
self.classifier = DomainClassifier()
self.g_optimizer = torch.optim.Adam(self.generator.parameters(),
self.g_lr,
[self.beta1, self.beta2])
self.d_optimizer = torch.optim.Adam(self.discriminator.parameters(),
self.d_lr,
[self.beta1, self.beta2])
self.c_optimizer = torch.optim.Adam(self.discriminator.parameters(),
self.c_lr,
[self.beta1, self.beta2])
self.print_network(self.generator, 'Generator')
self.print_network(self.discriminator, 'Discriminator')
self.print_network(self.classifier, 'Domain Classifier')
self.generator.to(self.device)
self.discriminator.to(self.device)
self.classifier.to(self.device)
def print_network(self, model, name):
"""Print out the network information."""
num_params = 0
for p in model.parameters():
num_params += p.numel()
print(model)
print(name)
print("The number of parameters: {}".format(num_params))
def restore_model(self, resume_iters):
"""Restore the trained generator and discriminator."""
print(
'Loading the trained models from step {}...'.format(resume_iters))
g_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(resume_iters))
d_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(resume_iters))
c_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(resume_iters))
self.generator.load_state_dict(
torch.load(g_path, map_location=lambda storage, loc: storage))
self.discriminator.load_state_dict(
torch.load(d_path, map_location=lambda storage, loc: storage))
self.classifier.load_state_dict(
torch.load(c_path, map_location=lambda storage, loc: storage))
def build_tensorboard(self):
"""Build a tensorboard logger."""
from logger import Logger
self.logger = Logger(self.log_dir)
def update_lr(self, g_lr, d_lr, c_lr):
"""Decay learning rates of the generator and discriminator."""
for param_group in self.g_optimizer.param_groups:
param_group['lr'] = g_lr
for param_group in self.d_optimizer.param_groups:
param_group['lr'] = d_lr
for param_group in self.c_optimizer.param_groups:
param_group['lr'] = c_lr
def reset_grad(self):
"""Reset the gradientgradient buffers."""
self.g_optimizer.zero_grad()
self.d_optimizer.zero_grad()
self.c_optimizer.zero_grad()
def denorm(self, x):
"""Convert the range from [-1, 1] to [0, 1]."""
out = (x + 1) / 2
return out.clamp_(0, 1)
def gradient_penalty(self, y, x):
"""Compute gradient penalty: (L2_norm(dy/dx) - 1)**2."""
weight = torch.ones(y.size()).to(self.device)
dydx = torch.autograd.grad(outputs=y,
inputs=x,
grad_outputs=weight,
retain_graph=True,
create_graph=True,
only_inputs=True)[0]
dydx = dydx.view(dydx.size(0), -1)
dydx_l2norm = torch.sqrt(torch.sum(dydx**2, dim=1))
return torch.mean((dydx_l2norm - 1)**2)
def label2onehot(self, labels, dim):
"""Convert label indices to one-hot vectors."""
batch_size = labels.size(0)
out = torch.zeros(batch_size, dim)
out[np.arange(batch_size), labels.long()] = 1
return out
def sample_spk_c(self, size):
spk_c = np.random.randint(0, self.num_speakers, size=size)
spk_c_cat = to_categorical(spk_c, self.num_speakers)
return torch.LongTensor(spk_c), torch.FloatTensor(spk_c_cat)
def classification_loss(self, logit, target):
"""Compute softmax cross entropy loss."""
return F.cross_entropy(logit, target)
def load_wav(self, wavfile, sr=16000):
wav, _ = librosa.load(wavfile, sr=sr, mono=True)
return wav_padding(wav, sr=16000, frame_period=5, multiple=4)
def train(self):
"""Train StarGAN."""
# Set data loader.
train_loader = self.train_loader
data_iter = iter(train_loader)
# Read a batch of testdata
test_wavfiles = self.test_loader.get_batch_test_data(batch_size=4)
test_wavs = [self.load_wav(wavfile) for wavfile in test_wavfiles]
# Determine whether do copysynthesize when first do training-time conversion test.
cpsyn_flag = [True, False][0]
# f0, timeaxis, sp, ap = world_decompose(wav = wav, fs = sampling_rate, frame_period = frame_period)
# Learning rate cache for decaying.
g_lr = self.g_lr
d_lr = self.d_lr
c_lr = self.c_lr
# Start training from scratch or resume training.
start_iters = 0
if self.resume_iters:
print("resuming step %d ..." % self.resume_iters)
start_iters = self.resume_iters
self.restore_model(self.resume_iters)
# Start training.
print('Start training...')
start_time = time.time()
for i in range(start_iters, self.num_iters):
# =================================================================================== #
# 1. Preprocess input data #
# =================================================================================== #
# Fetch labels.
try:
mc_real, spk_label_org, spk_c_org = next(data_iter)
except:
data_iter = iter(train_loader)
mc_real, spk_label_org, spk_c_org = next(data_iter)
mc_real.unsqueeze_(1) # (B, D, T) -> (B, 1, D, T) for conv2d
# Generate target domain labels randomly.
# spk_label_trg: int, spk_c_trg:one-hot representation
spk_label_trg, spk_c_trg = self.sample_spk_c(mc_real.size(0))
mc_real = mc_real.to(self.device) # Input mc.
spk_label_org = spk_label_org.to(
self.device) # Original spk labels.
spk_c_org = spk_c_org.to(self.device) # Original spk one-hot.
spk_label_trg = spk_label_trg.to(self.device) # Target spk labels.
spk_c_trg = spk_c_trg.to(self.device) # Target spk one-hot.
# =================================================================================== #
# 2. Train the Domain Classifier #
# =================================================================================== #
# Compute real classification loss.
cls_real = self.classifier(mc_real)
cls_loss = self.classification_loss(cls_real, spk_label_org)
# Backwards and optimize
self.reset_grad()
cls_loss.backward()
self.c_optimizer.step()
# Logging.
loss = {}
loss['C/c_loss'] = cls_loss.item()
# =================================================================================== #
# 3. Train the Discriminator #
# =================================================================================== #
# Compute loss with real mc feats.
d_out_src = self.discriminator(mc_real, spk_c_org)
mc_fake = self.generator(mc_real, spk_c_trg)
d_out_fake = self.discriminator(mc_fake.detach(), spk_c_trg)
d_loss = F.binary_cross_entropy_with_logits(d_out_fake, torch.zeros_like(d_out_fake, dtype=torch.float)) + \
F.binary_cross_entropy_with_logits(d_out_src, torch.ones_like(d_out_src, dtype=torch.float))
# Compute classification loss.
d_out_cls = self.classifier(mc_fake)
d_loss_cls = self.classification_loss(d_out_cls, spk_label_trg)
# Compute loss for gradient penalty.
alpha = torch.rand(mc_real.size(0), 1, 1, 1).to(self.device)
x_hat = (alpha * mc_real.data +
(1 - alpha) * mc_fake.data).requires_grad_(True)
d_out_src = self.discriminator(x_hat, spk_c_trg)
d_loss_gp = self.gradient_penalty(d_out_src, x_hat)
# Backward and optimize.
d_loss = d_loss + self.lambda_cls * d_loss_cls + self.lambda_gp * d_loss_gp
self.reset_grad()
d_loss.backward()
self.d_optimizer.step()
# Logging.
loss['D/loss_gp'] = d_loss_gp.item()
loss['D/loss'] = d_loss.item()
# =================================================================================== #
# 4. Train the generator #
# =================================================================================== #
if (i + 1) % self.n_critic == 0:
# Original-to-target domain.
mc_fake = self.generator(mc_real, spk_c_trg)
g_out_src = self.discriminator(mc_fake, spk_c_trg)
g_loss_fake = -torch.mean(g_out_src)
# Classification loss.
g_out_cls = self.classifier(mc_fake)
g_loss_cls = self.classification_loss(g_out_cls, spk_label_trg)
# Target-to-original domain. Cycle-consistent.
mc_reconst = self.generator(mc_fake, spk_c_org)
g_loss_rec = torch.mean(torch.abs(mc_real - mc_reconst))
# Original-to-original, Id mapping loss. Mapping
mc_fake_id = self.generator(mc_real, spk_c_org)
g_loss_id = torch.mean(torch.abs(mc_real - mc_fake_id))
# Backward and optimize.
g_loss = g_loss_fake \
+ self.lambda_rec * g_loss_rec \
+ self.lambda_cls * g_loss_cls \
+ self.lambda_id * g_loss_id
self.reset_grad()
g_loss.backward()
self.g_optimizer.step()
# Logging.
loss['G/loss_fake'] = g_loss_fake.item()
loss['G/loss_rec'] = g_loss_rec.item()
loss['G/loss_cls'] = g_loss_cls.item()
loss['G/loss'] = g_loss.item()
# =================================================================================== #
# 5. Miscellaneous #
# =================================================================================== #
# Print out training information.
if (i + 1) % self.log_step == 0:
et = time.time() - start_time
et = str(datetime.timedelta(seconds=et))[:-7]
log = "Elapsed [{}], Iteration [{}/{}]".format(
et, i + 1, self.num_iters)
for tag, value in loss.items():
log += ", {}: {:.4f}".format(tag, value)
print(log)
if self.use_tensorboard:
for tag, value in loss.items():
self.logger.scalar_summary(tag, value, i + 1)
if (i + 1) % self.sample_step == 0:
sampling_rate = 16000
num_mcep = 36
frame_period = 5
with torch.no_grad():
for idx, wav in tqdm(enumerate(test_wavs)):
wav_name = basename(test_wavfiles[idx])
# print(wav_name)
f0, timeaxis, sp, ap = world_decompose(
wav=wav,
fs=sampling_rate,
frame_period=frame_period)
f0_converted = pitch_conversion(
f0=f0,
mean_log_src=self.test_loader.logf0s_mean_src,
std_log_src=self.test_loader.logf0s_std_src,
mean_log_target=self.test_loader.logf0s_mean_trg,
std_log_target=self.test_loader.logf0s_std_trg)
coded_sp = world_encode_spectral_envelop(
sp=sp, fs=sampling_rate, dim=num_mcep)
coded_sp_norm = (coded_sp -
self.test_loader.mcep_mean_src
) / self.test_loader.mcep_std_src
coded_sp_norm_tensor = torch.FloatTensor(
coded_sp_norm.T).unsqueeze_(0).unsqueeze_(1).to(
self.device)
conds = torch.FloatTensor(
self.test_loader.spk_c_trg).to(self.device)
# print(conds.size())
coded_sp_converted_norm = self.generator(
coded_sp_norm_tensor, conds).data.cpu().numpy()
coded_sp_converted = np.squeeze(
coded_sp_converted_norm
).T * self.test_loader.mcep_std_trg + self.test_loader.mcep_mean_trg
coded_sp_converted = np.ascontiguousarray(
coded_sp_converted)
# decoded_sp_converted = world_decode_spectral_envelop(coded_sp = coded_sp_converted, fs = sampling_rate)
wav_transformed = world_speech_synthesis(
f0=f0_converted,
coded_sp=coded_sp_converted,
ap=ap,
fs=sampling_rate,
frame_period=frame_period)
librosa.output.write_wav(
join(
self.sample_dir,
str(i + 1) + '-' + wav_name.split('.')[0] +
'-vcto-{}'.format(self.test_loader.trg_spk) +
'.wav'), wav_transformed, sampling_rate)
if cpsyn_flag:
wav_cpsyn = world_speech_synthesis(
f0=f0,
coded_sp=coded_sp,
ap=ap,
fs=sampling_rate,
frame_period=frame_period)
librosa.output.write_wav(
join(self.sample_dir, 'cpsyn-' + wav_name),
wav_cpsyn, sampling_rate)
cpsyn_flag = False
# Save model checkpoints.
if (i + 1) % self.model_save_step == 0:
g_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(i + 1))
d_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(i + 1))
c_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(i + 1))
torch.save(self.generator.state_dict(), g_path)
torch.save(self.discriminator.state_dict(), d_path)
torch.save(self.classifier.state_dict(), c_path)
print('Saved model checkpoints into {}...'.format(
self.model_save_dir))
# Decay learning rates.
if (i + 1) % self.lr_update_step == 0 and (i + 1) > (
self.num_iters - self.num_iters_decay):
g_lr -= (self.g_lr / float(self.num_iters_decay))
d_lr -= (self.d_lr / float(self.num_iters_decay))
c_lr -= (self.c_lr / float(self.num_iters_decay))
self.update_lr(g_lr, d_lr, c_lr)
print('Decayed learning rates, g_lr: {}, d_lr: {}, c_lr: {}.'.
format(g_lr, d_lr, c_lr))
class Solver(object):
"""docstring for Solver."""
def __init__(self, data_loader, config):
self.config = config
self.data_loader = data_loader
# Model configurations.
self.lambda_cycle = config.lambda_cycle
self.lambda_cls = config.lambda_cls
self.lambda_identity = config.lambda_identity
# Training configurations.
self.data_dir = config.data_dir
self.test_dir = config.test_dir
self.batch_size = config.batch_size
self.num_iters = config.num_iters
self.num_iters_decay = config.num_iters_decay
self.g_lr = config.g_lr
self.d_lr = config.d_lr
self.c_lr = config.c_lr
self.n_critic = config.n_critic
self.beta1 = config.beta1
self.beta2 = config.beta2
self.resume_iters = config.resume_iters
# Test configurations.
self.test_iters = config.test_iters
self.trg_speaker = ast.literal_eval(config.trg_speaker)
self.src_speaker = config.src_speaker
# Miscellaneous.
self.use_tensorboard = config.use_tensorboard
self.device = torch.device(
'cuda:0' if torch.cuda.is_available() else 'cpu')
self.spk_enc = LabelBinarizer().fit(speakers)
# Directories.
self.log_dir = config.log_dir
self.sample_dir = config.sample_dir
self.model_save_dir = config.model_save_dir
self.result_dir = config.result_dir
# Step size.
self.log_step = config.log_step
self.sample_step = config.sample_step
self.model_save_step = config.model_save_step
self.lr_update_step = config.lr_update_step
# Build the model and tensorboard.
self.build_model()
if self.use_tensorboard:
self.build_tensorboard()
def build_model(self):
self.G = Generator()
self.D = Discriminator()
self.C = DomainClassifier()
self.g_optimizer = torch.optim.Adam(self.G.parameters(), self.g_lr,
[self.beta1, self.beta2])
self.d_optimizer = torch.optim.Adam(self.D.parameters(), self.d_lr,
[self.beta1, self.beta2])
self.c_optimizer = torch.optim.Adam(self.C.parameters(), self.c_lr,
[self.beta1, self.beta2])
self.print_network(self.G, 'G')
self.print_network(self.D, 'D')
self.print_network(self.C, 'C')
self.G.to(self.device)
self.D.to(self.device)
self.C.to(self.device)
def print_network(self, model, name):
"""Print out the network information."""
num_params = 0
for p in model.parameters():
num_params += p.numel()
print(model)
print(name)
print("The number of parameters: {}".format(num_params))
def build_tensorboard(self):
"""Build a tensorboard logger."""
from logger import Logger
self.logger = Logger(self.log_dir)
def update_lr(self, g_lr, d_lr, c_lr):
"""Decay learning rates of the generator and discriminator and classifier."""
for param_group in self.g_optimizer.param_groups:
param_group['lr'] = g_lr
for param_group in self.d_optimizer.param_groups:
param_group['lr'] = d_lr
for param_group in self.c_optimizer.param_groups:
param_group['lr'] = c_lr
def train(self):
# Learning rate cache for decaying.
g_lr = self.g_lr
d_lr = self.d_lr
c_lr = self.c_lr
start_iters = 0
if self.resume_iters:
pass
norm = Normalizer()
data_iter = iter(self.data_loader)
print('Start training......')
start_time = datetime.now()
for i in range(start_iters, self.num_iters):
# =================================================================================== #
# 1. Preprocess input data #
# =================================================================================== #
# Fetch real images and labels.
try:
x_real, speaker_idx_org, label_org = next(data_iter)
except:
data_iter = iter(self.data_loader)
x_real, speaker_idx_org, label_org = next(data_iter)
# Generate target domain labels randomly.
rand_idx = torch.randperm(label_org.size(0))
label_trg = label_org[rand_idx]
speaker_idx_trg = speaker_idx_org[rand_idx]
x_real = x_real.to(self.device) # Input images.
label_org = label_org.to(
self.device) # Original domain one-hot labels.
label_trg = label_trg.to(
self.device) # Target domain one-hot labels.
speaker_idx_org = speaker_idx_org.to(
self.device) # Original domain labels
speaker_idx_trg = speaker_idx_trg.to(
self.device) #Target domain labels
# =================================================================================== #
# 2. Train the discriminator #
# =================================================================================== #
# Compute loss with real audio frame.
CELoss = nn.CrossEntropyLoss()
cls_real = self.C(x_real)
cls_loss_real = CELoss(input=cls_real, target=speaker_idx_org)
self.reset_grad()
cls_loss_real.backward()
self.c_optimizer.step()
# Logging.
loss = {}
loss['C/C_loss'] = cls_loss_real.item()
out_r = self.D(x_real, label_org)
# Compute loss with fake audio frame.
x_fake = self.G(x_real, label_trg)
out_f = self.D(x_fake.detach(), label_trg)
d_loss_t = F.binary_cross_entropy_with_logits(input=out_f,target=torch.zeros_like(out_f, dtype=torch.float)) + \
F.binary_cross_entropy_with_logits(input=out_r, target=torch.ones_like(out_r, dtype=torch.float))
out_cls = self.C(x_fake)
d_loss_cls = CELoss(input=out_cls, target=speaker_idx_trg)
# Compute loss for gradient penalty.
alpha = torch.rand(x_real.size(0), 1, 1, 1).to(self.device)
x_hat = (alpha * x_real.data +
(1 - alpha) * x_fake.data).requires_grad_(True)
out_src = self.D(x_hat, label_trg)
d_loss_gp = self.gradient_penalty(out_src, x_hat)
d_loss = d_loss_t + self.lambda_cls * d_loss_cls + 5 * d_loss_gp
self.reset_grad()
d_loss.backward()
self.d_optimizer.step()
# loss['D/d_loss_t'] = d_loss_t.item()
# loss['D/loss_cls'] = d_loss_cls.item()
# loss['D/D_gp'] = d_loss_gp.item()
loss['D/D_loss'] = d_loss.item()
# =================================================================================== #
# 3. Train the generator #
# =================================================================================== #
if (i + 1) % self.n_critic == 0:
# Original-to-target domain.
x_fake = self.G(x_real, label_trg)
g_out_src = self.D(x_fake, label_trg)
g_loss_fake = F.binary_cross_entropy_with_logits(
input=g_out_src,
target=torch.ones_like(g_out_src, dtype=torch.float))
out_cls = self.C(x_fake)
g_loss_cls = CELoss(input=out_cls, target=speaker_idx_trg)
# Target-to-original domain.
x_reconst = self.G(x_fake, label_org)
g_loss_rec = F.l1_loss(x_reconst, x_real)
# Original-to-Original domain(identity).
x_fake_iden = self.G(x_real, label_org)
id_loss = F.l1_loss(x_fake_iden, x_real)
# Backward and optimize.
g_loss = g_loss_fake + self.lambda_cycle * g_loss_rec +\
self.lambda_cls * g_loss_cls + self.lambda_identity * id_loss
self.reset_grad()
g_loss.backward()
self.g_optimizer.step()
# Logging.
loss['G/loss_fake'] = g_loss_fake.item()
loss['G/loss_rec'] = g_loss_rec.item()
loss['G/loss_cls'] = g_loss_cls.item()
loss['G/loss_id'] = id_loss.item()
loss['G/g_loss'] = g_loss.item()
# =================================================================================== #
# 4. Miscellaneous #
# =================================================================================== #
# Print out training information.
if (i + 1) % self.log_step == 0:
et = datetime.now() - start_time
et = str(et)[:-7]
log = "Elapsed [{}], Iteration [{}/{}]".format(
et, i + 1, self.num_iters)
for tag, value in loss.items():
log += ", {}: {:.4f}".format(tag, value)
print(log)
if self.use_tensorboard:
for tag, value in loss.items():
self.logger.scalar_summary(tag, value, i + 1)
# Translate fixed images for debugging.
if (i + 1) % self.sample_step == 0:
with torch.no_grad():
d, speaker = TestSet(self.test_dir).test_data()
target = random.choice(
[x for x in speakers if x != speaker])
label_t = self.spk_enc.transform([target])[0]
label_t = np.asarray([label_t])
for filename, content in d.items():
f0 = content['f0']
ap = content['ap']
sp_norm_pad = self.pad_coded_sp(
content['coded_sp_norm'])
convert_result = []
for start_idx in range(
0, sp_norm_pad.shape[1] - FRAMES + 1, FRAMES):
one_seg = sp_norm_pad[:,
start_idx:start_idx + FRAMES]
one_seg = torch.FloatTensor(one_seg).to(
self.device)
one_seg = one_seg.view(1, 1, one_seg.size(0),
one_seg.size(1))
l = torch.FloatTensor(label_t)
one_seg = one_seg.to(self.device)
l = l.to(self.device)
one_set_return = self.G(one_seg,
l).data.cpu().numpy()
one_set_return = np.squeeze(one_set_return)
one_set_return = norm.backward_process(
one_set_return, target)
convert_result.append(one_set_return)
convert_con = np.concatenate(convert_result, axis=1)
convert_con = convert_con[:,
0:content['coded_sp_norm'].
shape[1]]
contigu = np.ascontiguousarray(convert_con.T,
dtype=np.float64)
decoded_sp = decode_spectral_envelope(contigu,
SAMPLE_RATE,
fft_size=FFTSIZE)
f0_converted = norm.pitch_conversion(
f0, speaker, target)
wav = synthesize(f0_converted, decoded_sp, ap,
SAMPLE_RATE)
name = f'{speaker}-{target}_iter{i+1}_{filename}'
path = os.path.join(self.sample_dir, name)
print(f'[save]:{path}')
librosa.output.write_wav(path, wav, SAMPLE_RATE)
# Save model checkpoints.
if (i + 1) % self.model_save_step == 0:
G_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(i + 1))
D_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(i + 1))
C_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(i + 1))
torch.save(self.G.state_dict(), G_path)
torch.save(self.D.state_dict(), D_path)
torch.save(self.C.state_dict(), C_path)
print('Saved model checkpoints into {}...'.format(
self.model_save_dir))
# Decay learning rates.
if (i + 1) % self.lr_update_step == 0 and (i + 1) > (
self.num_iters - self.num_iters_decay):
g_lr -= (self.g_lr / float(self.num_iters_decay))
d_lr -= (self.d_lr / float(self.num_iters_decay))
c_lr -= (self.c_lr / float(self.num_iters_decay))
self.update_lr(g_lr, d_lr, c_lr)
print('Decayed learning rates, g_lr: {}, d_lr: {}.'.format(
g_lr, d_lr))
def gradient_penalty(self, y, x):
"""Compute gradient penalty: (L2_norm(dy/dx) - 1)**2."""
weight = torch.ones(y.size()).to(self.device)
dydx = torch.autograd.grad(outputs=y,
inputs=x,
grad_outputs=weight,
retain_graph=True,
create_graph=True,
only_inputs=True)[0]
dydx = dydx.view(dydx.size(0), -1)
dydx_l2norm = torch.sqrt(torch.sum(dydx**2, dim=1))
return torch.mean((dydx_l2norm - 1)**2)
def reset_grad(self):
"""Reset the gradient buffers."""
self.g_optimizer.zero_grad()
self.d_optimizer.zero_grad()
self.c_optimizer.zero_grad()
def restore_model(self, resume_iters):
"""Restore the trained generator and discriminator."""
print(
'Loading the trained models from step {}...'.format(resume_iters))
G_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(resume_iters))
D_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(resume_iters))
C_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(resume_iters))
self.G.load_state_dict(
torch.load(G_path, map_location=lambda storage, loc: storage))
self.D.load_state_dict(
torch.load(D_path, map_location=lambda storage, loc: storage))
self.C.load_state_dict(
torch.load(C_path, map_location=lambda storage, loc: storage))
@staticmethod
def pad_coded_sp(coded_sp_norm):
f_len = coded_sp_norm.shape[1]
if f_len >= FRAMES:
pad_length = FRAMES - (f_len - (f_len // FRAMES) * FRAMES)
elif f_len < FRAMES:
pad_length = FRAMES - f_len
sp_norm_pad = np.hstack(
(coded_sp_norm, np.zeros((coded_sp_norm.shape[0], pad_length))))
return sp_norm_pad
def test(self):
"""Translate speech using StarGAN ."""
# Load the trained generator.
self.restore_model(self.test_iters)
norm = Normalizer()
# Set data loader.
d, speaker = TestSet(self.test_dir).test_data(self.src_speaker)
targets = self.trg_speaker
for target in targets:
print(target)
assert target in speakers
label_t = self.spk_enc.transform([target])[0]
label_t = np.asarray([label_t])
with torch.no_grad():
for filename, content in d.items():
f0 = content['f0']
ap = content['ap']
sp_norm_pad = self.pad_coded_sp(content['coded_sp_norm'])
convert_result = []
for start_idx in range(0,
sp_norm_pad.shape[1] - FRAMES + 1,
FRAMES):
one_seg = sp_norm_pad[:, start_idx:start_idx + FRAMES]
one_seg = torch.FloatTensor(one_seg).to(self.device)
one_seg = one_seg.view(1, 1, one_seg.size(0),
one_seg.size(1))
l = torch.FloatTensor(label_t)
one_seg = one_seg.to(self.device)
l = l.to(self.device)
one_set_return = self.G(one_seg, l).data.cpu().numpy()
one_set_return = np.squeeze(one_set_return)
one_set_return = norm.backward_process(
one_set_return, target)
convert_result.append(one_set_return)
convert_con = np.concatenate(convert_result, axis=1)
convert_con = convert_con[:, 0:content['coded_sp_norm'].
shape[1]]
contigu = np.ascontiguousarray(convert_con.T,
dtype=np.float64)
decoded_sp = decode_spectral_envelope(contigu,
SAMPLE_RATE,
fft_size=FFTSIZE)
f0_converted = norm.pitch_conversion(f0, speaker, target)
wav = synthesize(f0_converted, decoded_sp, ap, SAMPLE_RATE)
name = f'{speaker}-{target}_iter{self.test_iters}_{filename}'
path = os.path.join(self.result_dir, name)
print(f'[save]:{path}')
librosa.output.write_wav(path, wav, SAMPLE_RATE)
class Solver(object):
"""docstring for Solver."""
def __init__(self, data_loader, config):
self.config = config
self.data_loader = data_loader
# Model configurations.
self.lambda_cycle = config.lambda_cycle
self.lambda_cls = config.lambda_cls
self.lambda_identity = config.lambda_identity
self.sigma_d = config.sigma_d
# Training configurations.
self.data_dir = config.data_dir
self.test_dir = config.test_dir
self.batch_size = config.batch_size
self.num_iters = config.num_iters
self.num_iters_decay = config.num_iters_decay
self.g_lr = config.g_lr
self.d_lr = config.d_lr
self.c_lr = config.c_lr
self.n_critic = config.n_critic
self.beta1 = config.beta1
self.beta2 = config.beta2
self.resume_iters = config.resume_iters
# Test configurations.
self.test_iters = config.test_iters
self.trg_style = ast.literal_eval(config.trg_style)
self.src_style = config.src_style
# Miscellaneous.
self.use_tensorboard = config.use_tensorboard
self.device = torch.device(
'cuda:0' if torch.cuda.is_available() else 'cpu')
self.stls_enc = LabelBinarizer().fit(styles)
# Directories.
self.log_dir = config.log_dir
self.sample_dir = config.sample_dir
self.model_save_dir = config.model_save_dir
self.result_dir = config.result_dir
# Step size.
self.log_step = config.log_step
self.sample_step = config.sample_step
self.model_save_step = config.model_save_step
self.lr_update_step = config.lr_update_step
# Build the model and tensorboard.
self.build_model()
if self.use_tensorboard:
self.build_tensorboard()
def build_model(self):
self.G = Generator()
self.D = Discriminator()
self.C = DomainClassifier()
self.g_optimizer = torch.optim.Adam(self.G.parameters(), self.g_lr,
[self.beta1, self.beta2])
self.d_optimizer = torch.optim.Adam(self.D.parameters(), self.d_lr,
[self.beta1, self.beta2])
self.c_optimizer = torch.optim.Adam(self.C.parameters(), self.c_lr,
[self.beta1, self.beta2])
self.print_network(self.G, 'G')
self.print_network(self.D, 'D')
self.print_network(self.C, 'C')
self.G.to(self.device)
self.D.to(self.device)
self.C.to(self.device)
def print_network(self, model, name):
"""Print out the network information."""
num_params = 0
for p in model.parameters():
num_params += p.numel()
print(model)
print(name)
print("The number of parameters: {}".format(num_params))
def build_tensorboard(self):
"""Build a tensorboard logger."""
from logger import Logger
self.logger = Logger(self.log_dir)
def update_lr(self, g_lr, d_lr, c_lr):
"""Decay learning rates of the generator and discriminator and classifier."""
for param_group in self.g_optimizer.param_groups:
param_group['lr'] = g_lr
for param_group in self.d_optimizer.param_groups:
param_group['lr'] = d_lr
for param_group in self.c_optimizer.param_groups:
param_group['lr'] = c_lr
def train(self):
# Learning rate cache for decaying.
g_lr = self.g_lr
d_lr = self.d_lr
c_lr = self.c_lr
start_iters = 0
if self.resume_iters:
pass
#norm = Normalizer()
data_iter = iter(self.data_loader)
print('Start training......')
start_time = datetime.now()
for i in range(start_iters, self.num_iters):
# =================================================================================== #
# 1. Preprocess input data #
# =================================================================================== #
# Fetch real images and labels.
try:
x_real, style_idx_org, label_org = next(data_iter)
except:
data_iter = iter(self.data_loader)
x_real, style_idx_org, label_org = next(data_iter)
#generate gaussian noise for robustness improvement
gaussian_noise = self.sigma_d * torch.randn(x_real.size())
# Generate target domain labels randomly.
rand_idx = torch.randperm(label_org.size(0))
label_trg = label_org[rand_idx]
style_idx_trg = style_idx_org[rand_idx]
x_real = x_real.to(self.device) # Input images.
label_org = label_org.to(
self.device) # Original domain one-hot labels.
label_trg = label_trg.to(
self.device) # Target domain one-hot labels.
style_idx_org = style_idx_org.to(
self.device) # Original domain labels
style_idx_trg = style_idx_trg.to(
self.device) #Target domain labels
gaussian_noise = gaussian_noise.to(
self.device) #gaussian noise for discriminators
# =================================================================================== #
# 2. Train the discriminator #
# =================================================================================== #
# Compute loss with real audio frame.
CELoss = nn.CrossEntropyLoss()
cls_real = self.C(x_real)
cls_loss_real = CELoss(input=cls_real, target=style_idx_org)
self.reset_grad()
cls_loss_real.backward()
self.c_optimizer.step()
# Logging.
loss = {}
loss['C/C_loss'] = cls_loss_real.item()
out_r = self.D(x_real + gaussian_noise, label_org)
# Compute loss with fake audio frame.
x_fake = self.G(x_real, label_trg)
out_f = self.D(x_fake + gaussian_noise, label_trg)
d_loss_t = F.mse_loss(input=out_f,target=torch.zeros_like(out_f, dtype=torch.float)) + \
F.mse_loss(input=out_r, target=torch.ones_like(out_r, dtype=torch.float))
out_cls = self.C(x_fake)
d_loss_cls = CELoss(input=out_cls, target=style_idx_trg)
#Compute loss for gradient penalty.
#alpha = torch.rand(x_real.size(0), 1, 1, 1).to(self.device)
#x_hat = (alpha * x_real.data + (1 - alpha) * x_fake.data).requires_grad_(True)
#out_src = self.D(x_hat, label_trg)
#d_loss_gp = self.gradient_penalty(out_src, x_hat)
d_loss = d_loss_t + self.lambda_cls * d_loss_cls
#\+ 5*d_loss_gp
self.reset_grad()
d_loss.backward()
self.d_optimizer.step()
# loss['D/d_loss_t'] = d_loss_t.item()
# loss['D/loss_cls'] = d_loss_cls.item()
# loss['D/D_gp'] = d_loss_gp.item()
loss['D/D_loss'] = d_loss.item()
# =================================================================================== #
# 3. Train the generator #
# =================================================================================== #
if (i + 1) % self.n_critic == 0:
# Original-to-target domain.
x_fake = self.G(x_real, label_trg)
g_out_src = self.D(x_fake + gaussian_noise, label_trg)
g_loss_fake = F.mse_loss(input=g_out_src,
target=torch.ones_like(
g_out_src, dtype=torch.float))
out_cls = self.C(x_real)
g_loss_cls = CELoss(input=out_cls, target=style_idx_org)
# Target-to-original domain.
x_reconst = self.G(x_fake, label_org)
g_loss_rec = F.l1_loss(x_reconst, x_real)
# Original-to-Original domain(identity).
x_fake_iden = self.G(x_real, label_org)
id_loss = F.l1_loss(x_fake_iden, x_real)
# Backward and optimize.
g_loss = g_loss_fake + self.lambda_cycle * g_loss_rec +\
self.lambda_cls * g_loss_cls + self.lambda_identity * id_loss
self.reset_grad()
g_loss.backward()
self.g_optimizer.step()
# Logging.
loss['G/loss_fake'] = g_loss_fake.item()
loss['G/loss_rec'] = g_loss_rec.item()
loss['G/loss_cls'] = g_loss_cls.item()
loss['G/loss_id'] = id_loss.item()
loss['G/g_loss'] = g_loss.item()
# =================================================================================== #
# 4. Miscellaneous #
# =================================================================================== #
# Print out training information.
if (i + 1) % self.log_step == 0:
et = datetime.now() - start_time
et = str(et)[:-7]
log = "Elapsed [{}], Iteration [{}/{}]".format(
et, i + 1, self.num_iters)
for tag, value in loss.items():
log += ", {}: {:.4f}".format(tag, value)
print(log)
if self.use_tensorboard:
for tag, value in loss.items():
self.logger.scalar_summary(tag, value, i + 1)
# Translate fixed images for debugging.
if (i + 1) % self.sample_step == 0:
with torch.no_grad():
d, style = TestSet(self.test_dir).test_data()
label_o = self.stls_enc.transform([style])[0]
label_o = np.asarray([label_o])
target = random.choice([x for x in styles if x != style])
label_t = self.stls_enc.transform([target])[0]
label_t = np.asarray([label_t])
for filename, content in d.items():
filename = filename.split('.')[0]
one_seg = torch.FloatTensor(content).to(self.device)
one_seg = one_seg.view(1, one_seg.size(0),
one_seg.size(1),
one_seg.size(2))
l_t = torch.FloatTensor(label_t)
one_seg = one_seg.to(self.device)
l_t = l_t.to(self.device)
one_set_transfer = self.G(one_seg, l_t)
l_o = torch.FloatTensor(label_o)
l_o = l_o.to(self.device)
one_set_cycle = self.G(
one_set_transfer.to(self.device),
l_o).data.cpu().numpy()
one_set_transfer = one_set_transfer.data.cpu().numpy()
one_set_transfer_binary = to_binary(
one_set_transfer, 0.5)
one_set_cycle_binary = to_binary(one_set_cycle, 0.5)
one_set_transfer_binary = one_set_transfer_binary.reshape(
-1, one_set_transfer_binary.shape[2],
one_set_transfer_binary.shape[3],
one_set_transfer_binary.shape[1])
one_set_cycle_binary = one_set_cycle_binary.reshape(
-1, one_set_cycle_binary.shape[2],
one_set_cycle_binary.shape[3],
one_set_cycle_binary.shape[1])
print(one_set_transfer_binary.shape,
one_set_cycle_binary.shape)
name_origin = f'{style}-{target}_iter{i+1}_{filename}_origin'
name_transfer = f'{style}-{target}_iter{i+1}_{filename}_transfer'
name_cycle = f'{style}-{target}_iter{i+1}_{filename}_cycle'
path_samples_per_iter = os.path.join(
self.sample_dir, f'iter{i+1}')
if not os.path.exists(path_samples_per_iter):
os.makedirs(path_samples_per_iter)
path_origin = os.path.join(path_samples_per_iter,
name_origin)
path_transfer = os.path.join(path_samples_per_iter,
name_transfer)
path_cycle = os.path.join(path_samples_per_iter,
name_cycle)
print(
f'[save]:{path_origin},{path_transfer},{path_cycle}'
)
save_midis(
content.reshape(1, content.shape[1],
content.shape[2],
content.shape[0]),
'{}.mid'.format(path_origin))
save_midis(one_set_transfer_binary,
'{}.mid'.format(path_transfer))
save_midis(one_set_cycle_binary,
'{}.mid'.format(path_cycle))
# Save model checkpoints.
if (i + 1) % self.model_save_step == 0:
G_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(i + 1))
D_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(i + 1))
C_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(i + 1))
torch.save(self.G.state_dict(), G_path)
torch.save(self.D.state_dict(), D_path)
torch.save(self.C.state_dict(), C_path)
print('Saved model checkpoints into {}...'.format(
self.model_save_dir))
# Decay learning rates.
if (i + 1) % self.lr_update_step == 0 and (i + 1) > (
self.num_iters - self.num_iters_decay):
g_lr -= (self.g_lr / float(self.num_iters_decay))
d_lr -= (self.d_lr / float(self.num_iters_decay))
c_lr -= (self.c_lr / float(self.num_iters_decay))
self.update_lr(g_lr, d_lr, c_lr)
print('Decayed learning rates, g_lr: {}, d_lr: {}.'.format(
g_lr, d_lr))
def reset_grad(self):
"""Reset the gradient buffers."""
self.g_optimizer.zero_grad()
self.d_optimizer.zero_grad()
self.c_optimizer.zero_grad()
def restore_model(self, resume_iters):
"""Restore the trained generator and discriminator."""
print(
'Loading the trained models from step {}...'.format(resume_iters))
G_path = os.path.join(self.model_save_dir,
'{}-G.ckpt'.format(resume_iters))
D_path = os.path.join(self.model_save_dir,
'{}-D.ckpt'.format(resume_iters))
C_path = os.path.join(self.model_save_dir,
'{}-C.ckpt'.format(resume_iters))
self.G.load_state_dict(
torch.load(G_path, map_location=lambda storage, loc: storage))
self.D.load_state_dict(
torch.load(D_path, map_location=lambda storage, loc: storage))
self.C.load_state_dict(
torch.load(C_path, map_location=lambda storage, loc: storage))
def test(self):
"""Translate speech using StarGAN ."""
# Load the trained generator.
self.restore_model(self.test_iters)
# Set data loader.
d, style = TestSet(self.test_dir).test_data(self.src_style)
targets = self.trg_style
for target in targets:
print(target)
assert target in styles
label_t = self.stls_enc.transform([target])[0]
label_t = np.asarray([label_t])
with torch.no_grad():
for filename, content in d.items():
filename = filename.split('.')[0]
one_seg = torch.FloatTensor(content).to(self.device)
one_seg = one_seg.view(1, one_seg.size(0), one_seg.size(1),
one_seg.size(2))
l_t = torch.FloatTensor(label_t)
one_seg = one_seg.to(self.device)
l_t = l_t.to(self.device)
one_set_transfer = self.G(one_seg, l_t).cpu().numpy()
one_set_transfer_binary = to_binary(one_set_transfer, 0.5)
one_set_transfer_binary = one_set_transfer_binary.reshape(
-1, one_set_transfer_binary.shape[2],
one_set_transfer_binary.shape[3],
one_set_transfer_binary.shape[1])
name_origin = f'{style}-{target}_iter{i+1}_{filename}_origin'
name_transfer = f'{style}-{target}_iter{i+1}_{filename}_transfer'
path = os.path.join(self.result_dir, f'iter{i+1}')
path_origin = os.path.join(path, name_origin)
path_transfer = os.path.join(path, name_transfer)
print(f'[save]:{path_origin},{path_transfer}')
save_midis(
content.reshape(1, content.shape[1], content.shape[2],
content.shape[0]),
'{}.mid'.format(path_origin))
save_midis(one_set_transfer_binary,
'{}.mid'.format(path_transfer))