v1

VITSConditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech.md

+# VITS:Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech
+
+## 论文
+
+[VITS](https://arxiv.org/abs/2106.06103)
+
+## 模型结构
+
+在这项工作中，作者提出了一种并行端到端 TTS 方法，它生成的音频比目前的两阶段模型更自然。该方法采用了变异推理，并增加了归一化流和对抗训练过程，从而提高了生成模型的表现力。作者还提出了一种随机时长预测器，可根据输入文本合成具有不同节奏的语音。通过对潜在变量的不确定性建模和随机时长预测器，该的方法表达了自然的一对多关系，即一个文本输入可以用不同的音调和节奏以多种方式说话。在 LJ Speech（一个单一说话者数据集）上进行的人类主观评估（平均意见分或 MOS）表明，该方法优于最佳的公开 TTS 系统，其 MOS 可与标准答案相媲美。
+
+![image-20240829142644186](image-20240829142644186.png)
+
+## 环境配置
+
+### Docker（方法一）
+
+```
+#拉取镜像
+docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:2.1.0-ubuntu20.04-dtk24.04.1-py3.10
+#创建容器
+docker run --shm-size 80g --network=host --name=telechat --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined  -v /opt/hyhal:/opt/hyhal:ro -v <Host Path>:<Container Path> -it <Your Image ID> bash
+
+#安装依赖
+cd vits
+pip install -r requirements.txt -i https://pypi.mirrors.ustc.edu.cn/simple/
+apt-get install espeak
+cd monotonic_align
+python setup.py build_ext --inplace
+
+```
+
+### Anaconda（方法二）
+
+1、关于本项目 DCU 显卡所需的特殊深度学习库可从光合开发者社区下载安装： https://developer.hpccube.com/tool/
+
+```
+DTK驱动：dtk-24.04.1
+python：python3.10
+torch:2.1
+torchvision:0.16.0
+torchaudio:2.1.2
+```
+
+2、其它非特殊库参照 requirements.txt 安装
+
+```
+cd vits
+pip install -r requirements.txt -i https://pypi.mirrors.ustc.edu.cn/simple/
+apt-get install espeak
+cd monotonic_align
+python setup.py build_ext --inplace
+```
+
+## 训练步骤
+
+### 数据集下载与处理
+
+- 训练任务采用 LJ Speech-1.1 数据集，下载地址：https://data.keithito.com/data/speech/LJSpeech-1.1.tar.bz2
+
+- 下载完成后，在 vits 文件夹下创建连接：
+
+  ```
+  ln -s /path/to/LJSpeech-1.1/wavs DUMMY1
+  ```
+
+### 代码修改
+
+- STFT：
+
+  ```
+  vim /usr/local/lib/python3.10/site-packages/torch/functional.py
+  ##把第650行的 normalized, onesided, return_complex)
+  ##改为
+  normalized, onesided, return_complex=True)
+  ```
+
+### 单卡训练
+
+```
+bash run_single.sh
+```
+
+### 多卡训练
+
+```
+bash run_multi.sh
+```
+
+*注意，每次运行训练后，DUMMY1 目录下会多出许多 .pt 文件，如果再次训练的话会报错，所以需要每次训练后把所有的 .pt 文件给删除：
+
+```
+rm -rf DUMMY1/*.pt
+```
+
+## 源码仓库及问题反馈
+
+
+
+## 参考资料
+
+https://github.com/jaywalnut310/vits/tree/main
\ No newline at end of file

attentions.py

+import copy
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+import commons
+import modules
+from modules import LayerNorm
+   
+
+class Encoder(nn.Module):
+  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., window_size=4, **kwargs):
+    super().__init__()
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.window_size = window_size
+
+    self.drop = nn.Dropout(p_dropout)
+    self.attn_layers = nn.ModuleList()
+    self.norm_layers_1 = nn.ModuleList()
+    self.ffn_layers = nn.ModuleList()
+    self.norm_layers_2 = nn.ModuleList()
+    for i in range(self.n_layers):
+      self.attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, window_size=window_size))
+      self.norm_layers_1.append(LayerNorm(hidden_channels))
+      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout))
+      self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+  def forward(self, x, x_mask):
+    attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+    x = x * x_mask
+    for i in range(self.n_layers):
+      y = self.attn_layers[i](x, x, attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_1[i](x + y)
+
+      y = self.ffn_layers[i](x, x_mask)
+      y = self.drop(y)
+      x = self.norm_layers_2[i](x + y)
+    x = x * x_mask
+    return x
+
+
+class Decoder(nn.Module):
+  def __init__(self, hidden_channels, filter_channels, n_heads, n_layers, kernel_size=1, p_dropout=0., proximal_bias=False, proximal_init=True, **kwargs):
+    super().__init__()
+    self.hidden_channels = hidden_channels
+    self.filter_channels = filter_channels
+    self.n_heads = n_heads
+    self.n_layers = n_layers
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.proximal_bias = proximal_bias
+    self.proximal_init = proximal_init
+
+    self.drop = nn.Dropout(p_dropout)
+    self.self_attn_layers = nn.ModuleList()
+    self.norm_layers_0 = nn.ModuleList()
+    self.encdec_attn_layers = nn.ModuleList()
+    self.norm_layers_1 = nn.ModuleList()
+    self.ffn_layers = nn.ModuleList()
+    self.norm_layers_2 = nn.ModuleList()
+    for i in range(self.n_layers):
+      self.self_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout, proximal_bias=proximal_bias, proximal_init=proximal_init))
+      self.norm_layers_0.append(LayerNorm(hidden_channels))
+      self.encdec_attn_layers.append(MultiHeadAttention(hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout))
+      self.norm_layers_1.append(LayerNorm(hidden_channels))
+      self.ffn_layers.append(FFN(hidden_channels, hidden_channels, filter_channels, kernel_size, p_dropout=p_dropout, causal=True))
+      self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+  def forward(self, x, x_mask, h, h_mask):
+    """
+    x: decoder input
+    h: encoder output
+    """
+    self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(device=x.device, dtype=x.dtype)
+    encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+    x = x * x_mask
+    for i in range(self.n_layers):
+      y = self.self_attn_layers[i](x, x, self_attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_0[i](x + y)
+
+      y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+      y = self.drop(y)
+      x = self.norm_layers_1[i](x + y)
+      
+      y = self.ffn_layers[i](x, x_mask)
+      y = self.drop(y)
+      x = self.norm_layers_2[i](x + y)
+    x = x * x_mask
+    return x
+
+
+class MultiHeadAttention(nn.Module):
+  def __init__(self, channels, out_channels, n_heads, p_dropout=0., window_size=None, heads_share=True, block_length=None, proximal_bias=False, proximal_init=False):
+    super().__init__()
+    assert channels % n_heads == 0
+
+    self.channels = channels
+    self.out_channels = out_channels
+    self.n_heads = n_heads
+    self.p_dropout = p_dropout
+    self.window_size = window_size
+    self.heads_share = heads_share
+    self.block_length = block_length
+    self.proximal_bias = proximal_bias
+    self.proximal_init = proximal_init
+    self.attn = None
+
+    self.k_channels = channels // n_heads
+    self.conv_q = nn.Conv1d(channels, channels, 1)
+    self.conv_k = nn.Conv1d(channels, channels, 1)
+    self.conv_v = nn.Conv1d(channels, channels, 1)
+    self.conv_o = nn.Conv1d(channels, out_channels, 1)
+    self.drop = nn.Dropout(p_dropout)
+
+    if window_size is not None:
+      n_heads_rel = 1 if heads_share else n_heads
+      rel_stddev = self.k_channels**-0.5
+      self.emb_rel_k = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
+      self.emb_rel_v = nn.Parameter(torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels) * rel_stddev)
+
+    nn.init.xavier_uniform_(self.conv_q.weight)
+    nn.init.xavier_uniform_(self.conv_k.weight)
+    nn.init.xavier_uniform_(self.conv_v.weight)
+    if proximal_init:
+      with torch.no_grad():
+        self.conv_k.weight.copy_(self.conv_q.weight)
+        self.conv_k.bias.copy_(self.conv_q.bias)
+      
+  def forward(self, x, c, attn_mask=None):
+    q = self.conv_q(x)
+    k = self.conv_k(c)
+    v = self.conv_v(c)
+    
+    x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+    x = self.conv_o(x)
+    return x
+
+  def attention(self, query, key, value, mask=None):
+    # reshape [b, d, t] -> [b, n_h, t, d_k]
+    b, d, t_s, t_t = (*key.size(), query.size(2))
+    query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3).contiguous()
+    key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3).contiguous()
+    value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3).contiguous()
+
+    scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1).contiguous())
+    if self.window_size is not None:
+      assert t_s == t_t, "Relative attention is only available for self-attention."
+      key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+      rel_logits = self._matmul_with_relative_keys(query /math.sqrt(self.k_channels), key_relative_embeddings)
+      scores_local = self._relative_position_to_absolute_position(rel_logits)
+      scores = scores + scores_local
+    if self.proximal_bias:
+      assert t_s == t_t, "Proximal bias is only available for self-attention."
+      scores = scores + self._attention_bias_proximal(t_s).to(device=scores.device, dtype=scores.dtype)
+    if mask is not None:
+      scores = scores.masked_fill(mask == 0, -1e4)
+      if self.block_length is not None:
+        assert t_s == t_t, "Local attention is only available for self-attention."
+        block_mask = torch.ones_like(scores).triu(-self.block_length).tril(self.block_length)
+        scores = scores.masked_fill(block_mask == 0, -1e4)
+    p_attn = F.softmax(scores, dim=-1) # [b, n_h, t_t, t_s]
+    p_attn = self.drop(p_attn)
+    output = torch.matmul(p_attn, value)
+    if self.window_size is not None:
+      relative_weights = self._absolute_position_to_relative_position(p_attn)
+      value_relative_embeddings = self._get_relative_embeddings(self.emb_rel_v, t_s)
+      output = output + self._matmul_with_relative_values(relative_weights, value_relative_embeddings)
+    output = output.transpose(2, 3).contiguous().view(b, d, t_t).contiguous() # [b, n_h, t_t, d_k] -> [b, d, t_t]
+    return output, p_attn
+
+  def _matmul_with_relative_values(self, x, y):
+    """
+    x: [b, h, l, m]
+    y: [h or 1, m, d]
+    ret: [b, h, l, d]
+    """
+    ret = torch.matmul(x, y.unsqueeze(0))
+    return ret
+
+  def _matmul_with_relative_keys(self, x, y):
+    """
+    x: [b, h, l, d]
+    y: [h or 1, m, d]
+    ret: [b, h, l, m]
+    """
+    ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1).contiguous())
+    return ret
+
+  def _get_relative_embeddings(self, relative_embeddings, length):
+    max_relative_position = 2 * self.window_size + 1
+    # Pad first before slice to avoid using cond ops.
+    pad_length = max(length - (self.window_size + 1), 0)
+    slice_start_position = max((self.window_size + 1) - length, 0)
+    slice_end_position = slice_start_position + 2 * length - 1
+    if pad_length > 0:
+      padded_relative_embeddings = F.pad(
+          relative_embeddings,
+          commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]))
+    else:
+      padded_relative_embeddings = relative_embeddings
+    used_relative_embeddings = padded_relative_embeddings[:,slice_start_position:slice_end_position]
+    return used_relative_embeddings
+
+  def _relative_position_to_absolute_position(self, x):
+    """
+    x: [b, h, l, 2*l-1]
+    ret: [b, h, l, l]
+    """
+    batch, heads, length, _ = x.size()
+    # Concat columns of pad to shift from relative to absolute indexing.
+    x = F.pad(x, commons.convert_pad_shape([[0,0],[0,0],[0,0],[0,1]]))
+
+    # Concat extra elements so to add up to shape (len+1, 2*len-1).
+    x_flat = x.view([batch, heads, length * 2 * length]).contiguous()
+    x_flat = F.pad(x_flat, commons.convert_pad_shape([[0,0],[0,0],[0,length-1]]))
+
+    # Reshape and slice out the padded elements.
+    x_final = x_flat.view([batch, heads, length+1, 2*length-1]).contiguous()[:, :, :length, length-1:]
+    return x_final
+
+  def _absolute_position_to_relative_position(self, x):
+    """
+    x: [b, h, l, l]
+    ret: [b, h, l, 2*l-1]
+    """
+    batch, heads, length, _ = x.size()
+    # padd along column
+    x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length-1]]))
+    x_flat = x.view([batch, heads, length**2 + length*(length -1)]).contiguous()
+    # add 0's in the beginning that will skew the elements after reshape
+    x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+    x_final = x_flat.view([batch, heads, length, 2*length]).contiguous()[:,:,:,1:]
+    return x_final
+
+  def _attention_bias_proximal(self, length):
+    """Bias for self-attention to encourage attention to close positions.
+    Args:
+      length: an integer scalar.
+    Returns:
+      a Tensor with shape [1, 1, length, length]
+    """
+    r = torch.arange(length, dtype=torch.float32)
+    diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+    return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+  def __init__(self, in_channels, out_channels, filter_channels, kernel_size, p_dropout=0., activation=None, causal=False):
+    super().__init__()
+    self.in_channels = in_channels
+    self.out_channels = out_channels
+    self.filter_channels = filter_channels
+    self.kernel_size = kernel_size
+    self.p_dropout = p_dropout
+    self.activation = activation
+    self.causal = causal
+
+    if causal:
+      self.padding = self._causal_padding
+    else:
+      self.padding = self._same_padding
+
+    self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+    self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+    self.drop = nn.Dropout(p_dropout)
+
+  def forward(self, x, x_mask):
+    x = self.conv_1(self.padding(x * x_mask))
+    if self.activation == "gelu":
+      x = x * torch.sigmoid(1.702 * x)
+    else:
+      x = torch.relu(x)
+    x = self.drop(x)
+    x = self.conv_2(self.padding(x * x_mask))
+    return x * x_mask
+  
+  def _causal_padding(self, x):
+    if self.kernel_size == 1:
+      return x
+    pad_l = self.kernel_size - 1
+    pad_r = 0
+    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+    x = F.pad(x, commons.convert_pad_shape(padding))
+    return x
+
+  def _same_padding(self, x):
+    if self.kernel_size == 1:
+      return x
+    pad_l = (self.kernel_size - 1) // 2
+    pad_r = self.kernel_size // 2
+    padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+    x = F.pad(x, commons.convert_pad_shape(padding))
+    return x

commons.py

+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+  classname = m.__class__.__name__
+  if classname.find("Conv") != -1:
+    m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+  return int((kernel_size*dilation - dilation)/2)
+
+
+def convert_pad_shape(pad_shape):
+  l = pad_shape[::-1]
+  pad_shape = [item for sublist in l for item in sublist]
+  return pad_shape
+
+
+def intersperse(lst, item):
+  result = [item] * (len(lst) * 2 + 1)
+  result[1::2] = lst
+  return result
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+  """KL(P||Q)"""
+  kl = (logs_q - logs_p) - 0.5
+  kl += 0.5 * (torch.exp(2. * logs_p) + ((m_p - m_q)**2)) * torch.exp(-2. * logs_q)
+  return kl
+
+
+def rand_gumbel(shape):
+  """Sample from the Gumbel distribution, protect from overflows."""
+  uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+  return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+  g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+  return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+  ret = torch.zeros_like(x[:, :, :segment_size])
+  for i in range(x.size(0)):
+    idx_str = ids_str[i]
+    idx_end = idx_str + segment_size
+    ret[i] = x[i, :, idx_str:idx_end]
+  return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+  b, d, t = x.size()
+  if x_lengths is None:
+    x_lengths = t
+  ids_str_max = x_lengths - segment_size + 1
+  ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+  ret = slice_segments(x, ids_str, segment_size)
+  return ret, ids_str
+
+
+def get_timing_signal_1d(
+    length, channels, min_timescale=1.0, max_timescale=1.0e4):
+  position = torch.arange(length, dtype=torch.float)
+  num_timescales = channels // 2
+  log_timescale_increment = (
+      math.log(float(max_timescale) / float(min_timescale)) /
+      (num_timescales - 1))
+  inv_timescales = min_timescale * torch.exp(
+      torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment)
+  scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+  signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+  signal = F.pad(signal, [0, 0, 0, channels % 2])
+  signal = signal.view(1, channels, length)
+  return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+  b, channels, length = x.size()
+  signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+  return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+  b, channels, length = x.size()
+  signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+  return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+  mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+  return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+  n_channels_int = n_channels[0]
+  in_act = input_a + input_b
+  t_act = torch.tanh(in_act[:, :n_channels_int, :])
+  s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+  acts = t_act * s_act
+  return acts
+
+
+def convert_pad_shape(pad_shape):
+  l = pad_shape[::-1]
+  pad_shape = [item for sublist in l for item in sublist]
+  return pad_shape
+
+
+def shift_1d(x):
+  x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+  return x
+
+
+def sequence_mask(length, max_length=None):
+  if max_length is None:
+    max_length = length.max()
+  x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+  return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+  """
+  duration: [b, 1, t_x]
+  mask: [b, 1, t_y, t_x]
+  """
+  device = duration.device
+  
+  b, _, t_y, t_x = mask.shape
+  cum_duration = torch.cumsum(duration, -1)
+  
+  cum_duration_flat = cum_duration.view(b * t_x)
+  path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+  path = path.view(b, t_x, t_y)
+  path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+  path = path.unsqueeze(1).transpose(2,3) * mask
+  return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+  if isinstance(parameters, torch.Tensor):
+    parameters = [parameters]
+  parameters = list(filter(lambda p: p.grad is not None, parameters))
+  norm_type = float(norm_type)
+  if clip_value is not None:
+    clip_value = float(clip_value)
+
+  total_norm = 0
+  for p in parameters:
+    param_norm = p.grad.data.norm(norm_type)
+    total_norm += param_norm.item() ** norm_type
+    if clip_value is not None:
+      p.grad.data.clamp_(min=-clip_value, max=clip_value)
+  total_norm = total_norm ** (1. / norm_type)
+  return total_norm

configs/ljs_base.json

+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 64,
+    "fp16_run": false,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/ljs_audio_text_train_filelist.txt.cleaned",
+    "validation_files":"filelists/ljs_audio_text_val_filelist.txt.cleaned",
+    "text_cleaners":["english_cleaners2"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 0,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false
+  }
+}

configs/ljs_nosdp.json

+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 64,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/ljs_audio_text_train_filelist.txt.cleaned",
+    "validation_files":"filelists/ljs_audio_text_val_filelist.txt.cleaned",
+    "text_cleaners":["english_cleaners2"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 0,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "use_sdp": false
+  }
+}

configs/vctk_base.json

+{
+  "train": {
+    "log_interval": 200,
+    "eval_interval": 1000,
+    "seed": 1234,
+    "epochs": 10000,
+    "learning_rate": 2e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 64,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 8192,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "training_files":"filelists/vctk_audio_sid_text_train_filelist.txt.cleaned",
+    "validation_files":"filelists/vctk_audio_sid_text_val_filelist.txt.cleaned",
+    "text_cleaners":["english_cleaners2"],
+    "max_wav_value": 32768.0,
+    "sampling_rate": 22050,
+    "filter_length": 1024,
+    "hop_length": 256,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null,
+    "add_blank": true,
+    "n_speakers": 109,
+    "cleaned_text": true
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0.1,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [8,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "n_layers_q": 3,
+    "use_spectral_norm": false,
+    "gin_channels": 256
+  }
+}

data_utils.py

+import time
+import os
+import random
+import numpy as np
+import torch
+import torch.utils.data
+
+import commons 
+from mel_processing import spectrogram_torch
+from utils import load_wav_to_torch, load_filepaths_and_text
+from text import text_to_sequence, cleaned_text_to_sequence
+
+
+class TextAudioLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+    def __init__(self, audiopaths_and_text, hparams):
+        self.audiopaths_and_text = load_filepaths_and_text(audiopaths_and_text)
+        self.text_cleaners  = hparams.text_cleaners
+        self.max_wav_value  = hparams.max_wav_value
+        self.sampling_rate  = hparams.sampling_rate
+        self.filter_length  = hparams.filter_length 
+        self.hop_length     = hparams.hop_length 
+        self.win_length     = hparams.win_length
+        self.sampling_rate  = hparams.sampling_rate 
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 190)
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_and_text)
+        self._filter()
+
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_and_text_new = []
+        lengths = []
+        for audiopath, text in self.audiopaths_and_text:
+            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
+                audiopaths_and_text_new.append([audiopath, text])
+                lengths.append(os.path.getsize(audiopath) // (2 * self.hop_length))
+        self.audiopaths_and_text = audiopaths_and_text_new
+        self.lengths = lengths
+
+    def get_audio_text_pair(self, audiopath_and_text):
+        # separate filename and text
+        audiopath, text = audiopath_and_text[0], audiopath_and_text[1]
+        text = self.get_text(text)
+        spec, wav = self.get_audio(audiopath)
+        return (text, spec, wav)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError("{} {} SR doesn't match target {} SR".format(
+                sampling_rate, self.sampling_rate))
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            spec = torch.load(spec_filename)
+        else:
+            spec = spectrogram_torch(audio_norm, self.filter_length,
+                self.sampling_rate, self.hop_length, self.win_length,
+                center=False)
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+        return spec, audio_norm
+
+    def get_text(self, text):
+        if self.cleaned_text:
+            text_norm = cleaned_text_to_sequence(text)
+        else:
+            text_norm = text_to_sequence(text, self.text_cleaners)
+        if self.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def __getitem__(self, index):
+        return self.get_audio_text_pair(self.audiopaths_and_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_and_text)
+
+
+class TextAudioCollate():
+    """ Zero-pads model inputs and targets
+    """
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text and aduio
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[1].size(1) for x in batch]),
+            dim=0, descending=True)
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, :text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, :spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, :wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+        if self.return_ids:
+            return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, ids_sorted_decreasing
+        return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths
+
+
+"""Multi speaker version"""
+class TextAudioSpeakerLoader(torch.utils.data.Dataset):
+    """
+        1) loads audio, speaker_id, text pairs
+        2) normalizes text and converts them to sequences of integers
+        3) computes spectrograms from audio files.
+    """
+    def __init__(self, audiopaths_sid_text, hparams):
+        self.audiopaths_sid_text = load_filepaths_and_text(audiopaths_sid_text)
+        self.text_cleaners = hparams.text_cleaners
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length  = hparams.filter_length
+        self.hop_length     = hparams.hop_length
+        self.win_length     = hparams.win_length
+        self.sampling_rate  = hparams.sampling_rate
+
+        self.cleaned_text = getattr(hparams, "cleaned_text", False)
+
+        self.add_blank = hparams.add_blank
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 190)
+
+        random.seed(1234)
+        random.shuffle(self.audiopaths_sid_text)
+        self._filter()
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+
+        audiopaths_sid_text_new = []
+        lengths = []
+        for audiopath, sid, text in self.audiopaths_sid_text:
+            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
+                audiopaths_sid_text_new.append([audiopath, sid, text])
+                lengths.append(os.path.getsize(audiopath) // (2 * self.hop_length))
+        self.audiopaths_sid_text = audiopaths_sid_text_new
+        self.lengths = lengths
+
+    def get_audio_text_speaker_pair(self, audiopath_sid_text):
+        # separate filename, speaker_id and text
+        audiopath, sid, text = audiopath_sid_text[0], audiopath_sid_text[1], audiopath_sid_text[2]
+        text = self.get_text(text)
+        spec, wav = self.get_audio(audiopath)
+        sid = self.get_sid(sid)
+        return (text, spec, wav, sid)
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError("{} {} SR doesn't match target {} SR".format(
+                sampling_rate, self.sampling_rate))
+        audio_norm = audio / self.max_wav_value
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            spec = torch.load(spec_filename)
+        else:
+            spec = spectrogram_torch(audio_norm, self.filter_length,
+                self.sampling_rate, self.hop_length, self.win_length,
+                center=False)
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename)
+        return spec, audio_norm
+
+    def get_text(self, text):
+        if self.cleaned_text:
+            text_norm = cleaned_text_to_sequence(text)
+        else:
+            text_norm = text_to_sequence(text, self.text_cleaners)
+        if self.add_blank:
+            text_norm = commons.intersperse(text_norm, 0)
+        text_norm = torch.LongTensor(text_norm)
+        return text_norm
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def __getitem__(self, index):
+        return self.get_audio_text_speaker_pair(self.audiopaths_sid_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_sid_text)
+
+
+class TextAudioSpeakerCollate():
+    """ Zero-pads model inputs and targets
+    """
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text, audio and speaker identities
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized, sid]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[1].size(1) for x in batch]),
+            dim=0, descending=True)
+
+        max_text_len = max([len(x[0]) for x in batch])
+        max_spec_len = max([x[1].size(1) for x in batch])
+        max_wav_len = max([x[2].size(1) for x in batch])
+
+        text_lengths = torch.LongTensor(len(batch))
+        spec_lengths = torch.LongTensor(len(batch))
+        wav_lengths = torch.LongTensor(len(batch))
+        sid = torch.LongTensor(len(batch))
+
+        text_padded = torch.LongTensor(len(batch), max_text_len)
+        spec_padded = torch.FloatTensor(len(batch), batch[0][1].size(0), max_spec_len)
+        wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
+        text_padded.zero_()
+        spec_padded.zero_()
+        wav_padded.zero_()
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            text = row[0]
+            text_padded[i, :text.size(0)] = text
+            text_lengths[i] = text.size(0)
+
+            spec = row[1]
+            spec_padded[i, :, :spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wav = row[2]
+            wav_padded[i, :, :wav.size(1)] = wav
+            wav_lengths[i] = wav.size(1)
+
+            sid[i] = row[3]
+
+        if self.return_ids:
+            return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, sid, ids_sorted_decreasing
+        return text_padded, text_lengths, spec_padded, spec_lengths, wav_padded, wav_lengths, sid
+
+
+class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
+    """
+    Maintain similar input lengths in a batch.
+    Length groups are specified by boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
+  
+    It removes samples which are not included in the boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
+    """
+    def __init__(self, dataset, batch_size, boundaries, num_replicas=None, rank=None, shuffle=True):
+        super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
+        self.lengths = dataset.lengths
+        self.batch_size = batch_size
+        self.boundaries = boundaries
+  
+        self.buckets, self.num_samples_per_bucket = self._create_buckets()
+        self.total_size = sum(self.num_samples_per_bucket)
+        self.num_samples = self.total_size // self.num_replicas
+  
+    def _create_buckets(self):
+        buckets = [[] for _ in range(len(self.boundaries) - 1)]
+        for i in range(len(self.lengths)):
+            length = self.lengths[i]
+            idx_bucket = self._bisect(length)
+            if idx_bucket != -1:
+                buckets[idx_bucket].append(i)
+  
+        for i in range(len(buckets) - 1, 0, -1):
+            if len(buckets[i]) == 0:
+                buckets.pop(i)
+                self.boundaries.pop(i+1)
+  
+        num_samples_per_bucket = []
+        for i in range(len(buckets)):
+            len_bucket = len(buckets[i])
+            total_batch_size = self.num_replicas * self.batch_size
+            rem = (total_batch_size - (len_bucket % total_batch_size)) % total_batch_size
+            num_samples_per_bucket.append(len_bucket + rem)
+        return buckets, num_samples_per_bucket
+  
+    def __iter__(self):
+      # deterministically shuffle based on epoch
+      g = torch.Generator()
+      g.manual_seed(self.epoch)
+  
+      indices = []
+      if self.shuffle:
+          for bucket in self.buckets:
+              indices.append(torch.randperm(len(bucket), generator=g).tolist())
+      else:
+          for bucket in self.buckets:
+              indices.append(list(range(len(bucket))))
+  
+      batches = []
+      for i in range(len(self.buckets)):
+          bucket = self.buckets[i]
+          len_bucket = len(bucket)
+          ids_bucket = indices[i]
+          num_samples_bucket = self.num_samples_per_bucket[i]
+  
+          # add extra samples to make it evenly divisible
+          rem = num_samples_bucket - len_bucket
+          ids_bucket = ids_bucket + ids_bucket * (rem // len_bucket) + ids_bucket[:(rem % len_bucket)]
+  
+          # subsample
+          ids_bucket = ids_bucket[self.rank::self.num_replicas]
+  
+          # batching
+          for j in range(len(ids_bucket) // self.batch_size):
+              batch = [bucket[idx] for idx in ids_bucket[j*self.batch_size:(j+1)*self.batch_size]]
+              batches.append(batch)
+  
+      if self.shuffle:
+          batch_ids = torch.randperm(len(batches), generator=g).tolist()
+          batches = [batches[i] for i in batch_ids]
+      self.batches = batches
+  
+      assert len(self.batches) * self.batch_size == self.num_samples
+      return iter(self.batches)
+  
+    def _bisect(self, x, lo=0, hi=None):
+      if hi is None:
+          hi = len(self.boundaries) - 1
+  
+      if hi > lo:
+          mid = (hi + lo) // 2
+          if self.boundaries[mid] < x and x <= self.boundaries[mid+1]:
+              return mid
+          elif x <= self.boundaries[mid]:
+              return self._bisect(x, lo, mid)
+          else:
+              return self._bisect(x, mid + 1, hi)
+      else:
+          return -1
+
+    def __len__(self):
+        return self.num_samples // self.batch_size