import copy import math import torch from torch import nn from torch.nn import functional as F import commons import modules import attentions import monotonic_align from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm from commons import init_weights, get_padding from pqmf import PQMF from stft import TorchSTFT, OnnxSTFT AVAILABLE_FLOW_TYPES = ["pre_conv", "pre_conv2", "fft", "mono_layer_inter_residual", "mono_layer_post_residual"] AVAILABLE_DURATION_DISCRIMINATOR_TYPES = {"dur_disc_1": "DurationDiscriminator", "dur_disc_2": "DurationDiscriminator2"} class StochasticDurationPredictor(nn.Module): def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, n_flows=4, gin_channels=0): super().__init__() filter_channels = in_channels # it needs to be removed from future version. self.in_channels = in_channels self.filter_channels = filter_channels self.kernel_size = kernel_size self.p_dropout = p_dropout self.n_flows = n_flows self.gin_channels = gin_channels self.log_flow = modules.Log() self.flows = nn.ModuleList() self.flows.append(modules.ElementwiseAffine(2)) for i in range(n_flows): self.flows.append(modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3)) self.flows.append(modules.Flip()) self.post_pre = nn.Conv1d(1, filter_channels, 1) self.post_proj = nn.Conv1d(filter_channels, filter_channels, 1) self.post_convs = modules.DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout) self.post_flows = nn.ModuleList() self.post_flows.append(modules.ElementwiseAffine(2)) for i in range(4): self.post_flows.append(modules.ConvFlow(2, filter_channels, kernel_size, n_layers=3)) self.post_flows.append(modules.Flip()) self.pre = nn.Conv1d(in_channels, filter_channels, 1) self.proj = nn.Conv1d(filter_channels, filter_channels, 1) self.convs = modules.DDSConv(filter_channels, kernel_size, n_layers=3, p_dropout=p_dropout) if gin_channels != 0: self.cond = nn.Conv1d(gin_channels, filter_channels, 1) def forward(self, x, x_mask, w=None, g=None, reverse=False, noise_scale=1.0): x = torch.detach(x) x = self.pre(x) if g is not None: g = torch.detach(g) x = x + self.cond(g) x = self.convs(x, x_mask) x = self.proj(x) * x_mask if not reverse: flows = self.flows assert w is not None logdet_tot_q = 0 h_w = self.post_pre(w) h_w = self.post_convs(h_w, x_mask) h_w = self.post_proj(h_w) * x_mask e_q = torch.randn(w.size(0), 2, w.size(2)).to(device=x.device, dtype=x.dtype) * x_mask z_q = e_q for flow in self.post_flows: z_q, logdet_q = flow(z_q, x_mask, g=(x + h_w)) logdet_tot_q += logdet_q z_u, z1 = torch.split(z_q, [1, 1], 1) u = torch.sigmoid(z_u) * x_mask z0 = (w - u) * x_mask logdet_tot_q += torch.sum((F.logsigmoid(z_u) + F.logsigmoid(-z_u)) * x_mask, [1, 2]) logq = torch.sum(-0.5 * (math.log(2 * math.pi) + (e_q ** 2)) * x_mask, [1, 2]) - logdet_tot_q logdet_tot = 0 z0, logdet = self.log_flow(z0, x_mask) logdet_tot += logdet z = torch.cat([z0, z1], 1) for flow in flows: z, logdet = flow(z, x_mask, g=x, reverse=reverse) logdet_tot = logdet_tot + logdet nll = torch.sum(0.5 * (math.log(2 * math.pi) + (z ** 2)) * x_mask, [1, 2]) - logdet_tot return nll + logq # [b] else: flows = list(reversed(self.flows)) flows = flows[:-2] + [flows[-1]] # remove a useless vflow z = torch.randn(x.size(0), 2, x.size(2)).to(device=x.device, dtype=x.dtype) * noise_scale for flow in flows: z = flow(z, x_mask, g=x, reverse=reverse) z0, z1 = torch.split(z, [1, 1], 1) logw = z0 return logw class DurationPredictor(nn.Module): def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0): super().__init__() self.in_channels = in_channels self.filter_channels = filter_channels self.kernel_size = kernel_size self.p_dropout = p_dropout self.gin_channels = gin_channels self.drop = nn.Dropout(p_dropout) self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size, padding=kernel_size // 2) self.norm_1 = modules.LayerNorm(filter_channels) self.conv_2 = nn.Conv1d(filter_channels, filter_channels, kernel_size, padding=kernel_size // 2) self.norm_2 = modules.LayerNorm(filter_channels) self.proj = nn.Conv1d(filter_channels, 1, 1) if gin_channels != 0: self.cond = nn.Conv1d(gin_channels, in_channels, 1) def forward(self, x, x_mask, g=None): x = torch.detach(x) if g is not None: g = torch.detach(g) x = x + self.cond(g) x = self.conv_1(x * x_mask) x = torch.relu(x) x = self.norm_1(x) x = self.drop(x) x = self.conv_2(x * x_mask) x = torch.relu(x) x = self.norm_2(x) x = self.drop(x) x = self.proj(x * x_mask) return x * x_mask class DurationDiscriminator(nn.Module): # vits2 # TODO : not using "spk conditioning" for now according to the paper. # Can be a better discriminator if we use it. def __init__(self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0): super().__init__() self.in_channels = in_channels self.filter_channels = filter_channels self.kernel_size = kernel_size self.p_dropout = p_dropout self.gin_channels = gin_channels self.drop = nn.Dropout(p_dropout) self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size, padding=kernel_size // 2) # self.norm_1 = modules.LayerNorm(filter_channels) self.conv_2 = nn.Conv1d(filter_channels, filter_channels, kernel_size, padding=kernel_size // 2) # self.norm_2 = modules.LayerNorm(filter_channels) self.dur_proj = nn.Conv1d(1, filter_channels, 1) self.pre_out_conv_1 = nn.Conv1d(2 * filter_channels, filter_channels, kernel_size, padding=kernel_size // 2) self.pre_out_norm_1 = modules.LayerNorm(filter_channels) self.pre_out_conv_2 = nn.Conv1d(filter_channels, filter_channels, kernel_size, padding=kernel_size // 2) self.pre_out_norm_2 = modules.LayerNorm(filter_channels) # if gin_channels != 0: # self.cond = nn.Conv1d(gin_channels, in_channels, 1) self.output_layer = nn.Sequential( nn.Linear(filter_channels, 1), nn.Sigmoid() ) def forward_probability(self, x, x_mask, dur, g=None): dur = self.dur_proj(dur) x = torch.cat([x, dur], dim=1) x = self.pre_out_conv_1(x * x_mask) # x = torch.relu(x) # x = self.pre_out_norm_1(x) # x = self.drop(x) x = self.pre_out_conv_2(x * x_mask) # x = torch.relu(x) # x = self.pre_out_norm_2(x) # x = self.drop(x) x = x * x_mask x = x.transpose(1, 2) output_prob = self.output_layer(x) return output_prob def forward(self, x, x_mask, dur_r, dur_hat, g=None): x = torch.detach(x) # if g is not None: # g = torch.detach(g) # x = x + self.cond(g) x = self.conv_1(x * x_mask) # x = torch.relu(x) # x = self.norm_1(x) # x = self.drop(x) x = self.conv_2(x * x_mask) # x = torch.relu(x) # x = self.norm_2(x) # x = self.drop(x) output_probs = [] for dur in [dur_r, dur_hat]: output_prob = self.forward_probability(x, x_mask, dur, g) output_probs.append(output_prob) return output_probs class DurationDiscriminator2(nn.Module): # vits2 - DurationDiscriminator2 # TODO : not using "spk conditioning" for now according to the paper. # Can be a better discriminator if we use it. def __init__( self, in_channels, filter_channels, kernel_size, p_dropout, gin_channels=0 ): super().__init__() self.in_channels = in_channels self.filter_channels = filter_channels self.kernel_size = kernel_size self.p_dropout = p_dropout self.gin_channels = gin_channels self.conv_1 = nn.Conv1d( in_channels, filter_channels, kernel_size, padding=kernel_size // 2 ) self.norm_1 = modules.LayerNorm(filter_channels) self.conv_2 = nn.Conv1d( filter_channels, filter_channels, kernel_size, padding=kernel_size // 2 ) self.norm_2 = modules.LayerNorm(filter_channels) self.dur_proj = nn.Conv1d(1, filter_channels, 1) self.pre_out_conv_1 = nn.Conv1d( 2 * filter_channels, filter_channels, kernel_size, padding=kernel_size // 2 ) self.pre_out_norm_1 = modules.LayerNorm(filter_channels) self.pre_out_conv_2 = nn.Conv1d( filter_channels, filter_channels, kernel_size, padding=kernel_size // 2 ) self.pre_out_norm_2 = modules.LayerNorm(filter_channels) # if gin_channels != 0: # self.cond = nn.Conv1d(gin_channels, in_channels, 1) self.output_layer = nn.Sequential(nn.Linear(filter_channels, 1), nn.Sigmoid()) def forward_probability(self, x, x_mask, dur, g=None): dur = self.dur_proj(dur) x = torch.cat([x, dur], dim=1) x = self.pre_out_conv_1(x * x_mask) x = torch.relu(x) x = self.pre_out_norm_1(x) x = self.pre_out_conv_2(x * x_mask) x = torch.relu(x) x = self.pre_out_norm_2(x) x = x * x_mask x = x.transpose(1, 2) output_prob = self.output_layer(x) return output_prob def forward(self, x, x_mask, dur_r, dur_hat, g=None): x = torch.detach(x) # if g is not None: # g = torch.detach(g) # x = x + self.cond(g) x = self.conv_1(x * x_mask) x = torch.relu(x) x = self.norm_1(x) x = self.conv_2(x * x_mask) x = torch.relu(x) x = self.norm_2(x) output_probs = [] for dur in [dur_r, dur_hat]: output_prob = self.forward_probability(x, x_mask, dur, g) output_probs.append([output_prob]) return output_probs class TextEncoder(nn.Module): def __init__(self, n_vocab, out_channels, hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout, gin_channels=0): super().__init__() self.n_vocab = n_vocab self.out_channels = out_channels 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.gin_channels = gin_channels self.emb = nn.Embedding(n_vocab, hidden_channels) nn.init.normal_(self.emb.weight, 0.0, hidden_channels ** -0.5) self.encoder = attentions.Encoder( hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout, gin_channels=self.gin_channels) self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1) def forward(self, x, x_lengths, g=None): x = self.emb(x) * math.sqrt(self.hidden_channels) # [b, t, h] x = torch.transpose(x, 1, -1) # [b, h, t] x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype) x = self.encoder(x * x_mask, x_mask, g=g) stats = self.proj(x) * x_mask m, logs = torch.split(stats, self.out_channels, dim=1) return x, m, logs, x_mask class ResidualCouplingTransformersLayer2(nn.Module): # vits2 def __init__( self, channels, hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=0, gin_channels=0, mean_only=False, ): assert channels % 2 == 0, "channels should be divisible by 2" super().__init__() self.channels = channels self.hidden_channels = hidden_channels self.kernel_size = kernel_size self.dilation_rate = dilation_rate self.n_layers = n_layers self.half_channels = channels // 2 self.mean_only = mean_only self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1) self.pre_transformer = attentions.Encoder( hidden_channels, hidden_channels, n_heads=2, n_layers=1, kernel_size=kernel_size, p_dropout=p_dropout, # window_size=None, ) self.enc = modules.WN( hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=p_dropout, gin_channels=gin_channels, ) self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1) self.post.weight.data.zero_() self.post.bias.data.zero_() def forward(self, x, x_mask, g=None, reverse=False): x0, x1 = torch.split(x, [self.half_channels] * 2, 1) h = self.pre(x0) * x_mask h = h + self.pre_transformer(h * x_mask, x_mask) # vits2 residual connection h = self.enc(h, x_mask, g=g) stats = self.post(h) * x_mask if not self.mean_only: m, logs = torch.split(stats, [self.half_channels] * 2, 1) else: m = stats logs = torch.zeros_like(m) if not reverse: x1 = m + x1 * torch.exp(logs) * x_mask x = torch.cat([x0, x1], 1) logdet = torch.sum(logs, [1, 2]) return x, logdet else: x1 = (x1 - m) * torch.exp(-logs) * x_mask x = torch.cat([x0, x1], 1) return x class ResidualCouplingTransformersLayer(nn.Module): # vits2 def __init__( self, channels, hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=0, gin_channels=0, mean_only=False, ): assert channels % 2 == 0, "channels should be divisible by 2" super().__init__() self.channels = channels self.hidden_channels = hidden_channels self.kernel_size = kernel_size self.dilation_rate = dilation_rate self.n_layers = n_layers self.half_channels = channels // 2 self.mean_only = mean_only # vits2 self.pre_transformer = attentions.Encoder( self.half_channels, self.half_channels, n_heads=2, n_layers=2, kernel_size=3, p_dropout=0.1, window_size=None ) self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1) self.enc = modules.WN( hidden_channels, kernel_size, dilation_rate, n_layers, p_dropout=p_dropout, gin_channels=gin_channels, ) # vits2 self.post_transformer = attentions.Encoder( self.hidden_channels, self.hidden_channels, n_heads=2, n_layers=2, kernel_size=3, p_dropout=0.1, window_size=None ) self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1) self.post.weight.data.zero_() self.post.bias.data.zero_() def forward(self, x, x_mask, g=None, reverse=False): x0, x1 = torch.split(x, [self.half_channels] * 2, 1) x0_ = self.pre_transformer(x0 * x_mask, x_mask) # vits2 x0_ = x0_ + x0 # vits2 residual connection h = self.pre(x0_) * x_mask # changed from x0 to x0_ to retain x0 for the flow h = self.enc(h, x_mask, g=g) # vits2 - (experimental;uncomment the following 2 line to use) # h_ = self.post_transformer(h, x_mask) # h = h + h_ #vits2 residual connection stats = self.post(h) * x_mask if not self.mean_only: m, logs = torch.split(stats, [self.half_channels] * 2, 1) else: m = stats logs = torch.zeros_like(m) if not reverse: x1 = m + x1 * torch.exp(logs) * x_mask x = torch.cat([x0, x1], 1) logdet = torch.sum(logs, [1, 2]) return x, logdet else: x1 = (x1 - m) * torch.exp(-logs) * x_mask x = torch.cat([x0, x1], 1) return x def remove_weight_norm(self): # ! self.enc.remove_weight_norm() class FFTransformerCouplingLayer(nn.Module): # vits2 def __init__(self, channels, hidden_channels, kernel_size, n_layers, n_heads, p_dropout=0, filter_channels=768, mean_only=False, gin_channels=0 ): assert channels % 2 == 0, "channels should be divisible by 2" super().__init__() self.channels = channels self.hidden_channels = hidden_channels self.kernel_size = kernel_size self.n_layers = n_layers self.half_channels = channels // 2 self.mean_only = mean_only self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1) self.enc = attentions.FFT( hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout, isflow=True, gin_channels=gin_channels ) self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1) self.post.weight.data.zero_() self.post.bias.data.zero_() def forward(self, x, x_mask, g=None, reverse=False): x0, x1 = torch.split(x, [self.half_channels] * 2, 1) h = self.pre(x0) * x_mask h_ = self.enc(h, x_mask, g=g) h = h_ + h stats = self.post(h) * x_mask if not self.mean_only: m, logs = torch.split(stats, [self.half_channels] * 2, 1) else: m = stats logs = torch.zeros_like(m) if not reverse: x1 = m + x1 * torch.exp(logs) * x_mask x = torch.cat([x0, x1], 1) logdet = torch.sum(logs, [1, 2]) return x, logdet else: x1 = (x1 - m) * torch.exp(-logs) * x_mask x = torch.cat([x0, x1], 1) return x class MonoTransformerFlowLayer(nn.Module): # vits2 def __init__( self, channels, hidden_channels, mean_only=False, residual_connection=False, # according to VITS-2 paper fig 1B set residual_connection=True ): assert channels % 2 == 0, "channels should be divisible by 2" super().__init__() self.channels = channels self.hidden_channels = hidden_channels self.half_channels = channels // 2 self.mean_only = mean_only self.residual_connection = residual_connection # vits2 self.pre_transformer = attentions.Encoder( self.half_channels, self.half_channels, n_heads=2, n_layers=2, kernel_size=3, p_dropout=0.1, window_size=None ) self.post = nn.Conv1d(self.half_channels, self.half_channels * (2 - mean_only), 1) self.post.weight.data.zero_() self.post.bias.data.zero_() def forward(self, x, x_mask, g=None, reverse=False): if self.residual_connection: if not reverse: x0, x1 = torch.split(x, [self.half_channels] * 2, 1) x0_ = x0 * x_mask x0_ = self.pre_transformer(x0, x_mask) # vits2 stats = self.post(x0_) * x_mask if not self.mean_only: m, logs = torch.split(stats, [self.half_channels] * 2, 1) else: m = stats logs = torch.zeros_like(m) x1 = m + x1 * torch.exp(logs) * x_mask x_ = torch.cat([x0, x1], 1) x = x + x_ logdet = torch.sum(torch.log(torch.exp(logs) + 1), [1, 2]) logdet = logdet + torch.log(torch.tensor(2)) * (x0.shape[1] * x0.shape[2]) return x, logdet else: x0, x1 = torch.split(x, [self.half_channels] * 2, 1) x0 = x0 / 2 x0_ = x0 * x_mask x0_ = self.pre_transformer(x0, x_mask) # vits2 stats = self.post(x0_) * x_mask if not self.mean_only: m, logs = torch.split(stats, [self.half_channels] * 2, 1) else: m = stats logs = torch.zeros_like(m) x1_ = ((x1 - m) / (1 + torch.exp(-logs))) * x_mask x = torch.cat([x0, x1_], 1) return x else: x0, x1 = torch.split(x, [self.half_channels] * 2, 1) x0_ = self.pre_transformer(x0 * x_mask, x_mask) # vits2 h = x0_ + x0 # vits2 stats = self.post(h) * x_mask if not self.mean_only: m, logs = torch.split(stats, [self.half_channels] * 2, 1) else: m = stats logs = torch.zeros_like(m) if not reverse: x1 = m + x1 * torch.exp(logs) * x_mask x = torch.cat([x0, x1], 1) logdet = torch.sum(logs, [1, 2]) return x, logdet else: x1 = (x1 - m) * torch.exp(-logs) * x_mask x = torch.cat([x0, x1], 1) return x class ResidualCouplingTransformersBlock(nn.Module): # vits2 def __init__(self, channels, hidden_channels, kernel_size, dilation_rate, n_layers, n_flows=4, gin_channels=0, use_transformer_flows=False, transformer_flow_type="pre_conv", ): super().__init__() self.channels = channels self.hidden_channels = hidden_channels self.kernel_size = kernel_size self.dilation_rate = dilation_rate self.n_layers = n_layers self.n_flows = n_flows self.gin_channels = gin_channels self.flows = nn.ModuleList() # TODO : clean up this mess if use_transformer_flows: if transformer_flow_type == "pre_conv": for i in range(n_flows): self.flows.append( ResidualCouplingTransformersLayer( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True ) ) self.flows.append(modules.Flip()) elif transformer_flow_type == "pre_conv2": for i in range(n_flows): self.flows.append( ResidualCouplingTransformersLayer2( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True, ) ) self.flows.append(modules.Flip()) elif transformer_flow_type == "fft": for i in range(n_flows): self.flows.append( FFTransformerCouplingLayer( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True ) ) self.flows.append(modules.Flip()) elif transformer_flow_type == "mono_layer_inter_residual": for i in range(n_flows): self.flows.append( modules.ResidualCouplingLayer( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True ) ) self.flows.append(modules.Flip()) self.flows.append( MonoTransformerFlowLayer( channels, hidden_channels, mean_only=True ) ) elif transformer_flow_type == "mono_layer_post_residual": for i in range(n_flows): self.flows.append( modules.ResidualCouplingLayer( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True, ) ) self.flows.append(modules.Flip()) self.flows.append( MonoTransformerFlowLayer( channels, hidden_channels, mean_only=True, residual_connection=True ) ) else: for i in range(n_flows): self.flows.append( modules.ResidualCouplingLayer( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True ) ) self.flows.append(modules.Flip()) def forward(self, x, x_mask, g=None, reverse=False): if not reverse: for flow in self.flows: x, _ = flow(x, x_mask, g=g, reverse=reverse) else: for flow in reversed(self.flows): x = flow(x, x_mask, g=g, reverse=reverse) return x def remove_weight_norm(self): # ! for i, l in enumerate(self.flows): if i % 2 == 0: l.remove_weight_norm() class ResidualCouplingBlock(nn.Module): def __init__(self, channels, hidden_channels, kernel_size, dilation_rate, n_layers, n_flows=4, gin_channels=0): super().__init__() self.channels = channels self.hidden_channels = hidden_channels self.kernel_size = kernel_size self.dilation_rate = dilation_rate self.n_layers = n_layers self.n_flows = n_flows self.gin_channels = gin_channels self.flows = nn.ModuleList() for i in range(n_flows): self.flows.append( modules.ResidualCouplingLayer( channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels, mean_only=True ) ) self.flows.append(modules.Flip()) def forward(self, x, x_mask, g=None, reverse=False): if not reverse: for flow in self.flows: x, _ = flow(x, x_mask, g=g, reverse=reverse) else: for flow in reversed(self.flows): x = flow(x, x_mask, g=g, reverse=reverse) return x def remove_weight_norm(self): # ! for i, l in enumerate(self.flows): if i % 2 == 0: l.remove_weight_norm() class PosteriorEncoder(nn.Module): def __init__(self, in_channels, out_channels, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=0): super().__init__() self.in_channels = in_channels self.out_channels = out_channels self.hidden_channels = hidden_channels self.kernel_size = kernel_size self.dilation_rate = dilation_rate self.n_layers = n_layers self.gin_channels = gin_channels self.pre = nn.Conv1d(in_channels, hidden_channels, 1) self.enc = modules.WN(hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=gin_channels) self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1) def forward(self, x, x_lengths, g=None): x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(x.dtype) x = self.pre(x) * x_mask x = self.enc(x, x_mask, g=g) stats = self.proj(x) * x_mask m, logs = torch.split(stats, self.out_channels, dim=1) z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask return z, m, logs, x_mask class Generator(torch.nn.Module): def __init__(self, initial_channel, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gin_channels=0): super(Generator, self).__init__() self.num_kernels = len(resblock_kernel_sizes) self.num_upsamples = len(upsample_rates) self.conv_pre = Conv1d(initial_channel, upsample_initial_channel, 7, 1, padding=3) resblock = modules.ResBlock1 if resblock == '1' else modules.ResBlock2 self.ups = nn.ModuleList() for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)): self.ups.append(weight_norm( ConvTranspose1d(upsample_initial_channel // (2 ** i), upsample_initial_channel // (2 ** (i + 1)), k, u, padding=(k - u) // 2))) self.resblocks = nn.ModuleList() for i in range(len(self.ups)): ch = upsample_initial_channel // (2 ** (i + 1)) for j, (k, d) in enumerate(zip(resblock_kernel_sizes, resblock_dilation_sizes)): self.resblocks.append(resblock(ch, k, d)) self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False) self.ups.apply(init_weights) if gin_channels != 0: self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1) def forward(self, x, g=None): x = self.conv_pre(x) if g is not None: x = x + self.cond(g) for i in range(self.num_upsamples): x = F.leaky_relu(x, modules.LRELU_SLOPE) x = self.ups[i](x) xs = None for j in range(self.num_kernels): if xs is None: xs = self.resblocks[i * self.num_kernels + j](x) else: xs += self.resblocks[i * self.num_kernels + j](x) x = xs / self.num_kernels x = F.leaky_relu(x) x = self.conv_post(x) x = torch.tanh(x) return x def remove_weight_norm(self): print('Removing weight norm...') for l in self.ups: remove_weight_norm(l) for l in self.resblocks: l.remove_weight_norm() class iSTFT_Generator(torch.nn.Module): def __init__(self, initial_channel, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, gin_channels=0, is_onnx=False): super(iSTFT_Generator, self).__init__() # self.h = h self.gen_istft_n_fft = gen_istft_n_fft self.gen_istft_hop_size = gen_istft_hop_size self.num_kernels = len(resblock_kernel_sizes) self.num_upsamples = len(upsample_rates) self.conv_pre = weight_norm(Conv1d(initial_channel, upsample_initial_channel, 7, 1, padding=3)) resblock = modules.ResBlock1 if resblock == '1' else modules.ResBlock2 self.ups = nn.ModuleList() for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)): self.ups.append(weight_norm( ConvTranspose1d(upsample_initial_channel // (2 ** i), upsample_initial_channel // (2 ** (i + 1)), k, u, padding=(k - u) // 2))) self.resblocks = nn.ModuleList() for i in range(len(self.ups)): ch = upsample_initial_channel // (2 ** (i + 1)) for j, (k, d) in enumerate(zip(resblock_kernel_sizes, resblock_dilation_sizes)): self.resblocks.append(resblock(ch, k, d)) self.post_n_fft = self.gen_istft_n_fft self.conv_post = weight_norm(Conv1d(ch, self.post_n_fft + 2, 7, 1, padding=3)) self.ups.apply(init_weights) self.conv_post.apply(init_weights) self.reflection_pad = torch.nn.ReflectionPad1d((1, 0)) ''' self.stft = TorchSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) ''' # - for onnx if is_onnx == True: self.stft = OnnxSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) else: self.stft = TorchSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) def forward(self, x, g=None): x = self.conv_pre(x) for i in range(self.num_upsamples): x = F.leaky_relu(x, modules.LRELU_SLOPE) x = self.ups[i](x) xs = None for j in range(self.num_kernels): if xs is None: xs = self.resblocks[i * self.num_kernels + j](x) else: xs += self.resblocks[i * self.num_kernels + j](x) x = xs / self.num_kernels x = F.leaky_relu(x) x = self.reflection_pad(x) x = self.conv_post(x) spec = torch.exp(x[:, :self.post_n_fft // 2 + 1, :]) phase = math.pi * torch.sin(x[:, self.post_n_fft // 2 + 1:, :]) out = self.stft.inverse(spec, phase).to(x.device) return out, None def remove_weight_norm(self): print('Removing weight norm...') for l in self.ups: remove_weight_norm(l) for l in self.resblocks: l.remove_weight_norm() remove_weight_norm(self.conv_pre) remove_weight_norm(self.conv_post) class Multiband_iSTFT_Generator(torch.nn.Module): # ! def __init__(self, initial_channel, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, subbands, gin_channels=0, is_onnx=False): super(Multiband_iSTFT_Generator, self).__init__() # self.h = h self.subbands = subbands self.num_kernels = len(resblock_kernel_sizes) self.num_upsamples = len(upsample_rates) self.conv_pre = weight_norm(Conv1d(initial_channel, upsample_initial_channel, 7, 1, padding=3)) resblock = modules.ResBlock1 if resblock == '1' else modules.ResBlock2 self.ups = nn.ModuleList() for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)): self.ups.append(weight_norm( ConvTranspose1d(upsample_initial_channel // (2 ** i), upsample_initial_channel // (2 ** (i + 1)), k, u, padding=(k - u) // 2))) self.resblocks = nn.ModuleList() for i in range(len(self.ups)): ch = upsample_initial_channel // (2 ** (i + 1)) for j, (k, d) in enumerate(zip(resblock_kernel_sizes, resblock_dilation_sizes)): self.resblocks.append(resblock(ch, k, d)) self.post_n_fft = gen_istft_n_fft self.ups.apply(init_weights) self.reflection_pad = torch.nn.ReflectionPad1d((1, 0)) self.reshape_pixelshuffle = [] self.subband_conv_post = weight_norm(Conv1d(ch, self.subbands * (self.post_n_fft + 2), 7, 1, padding=3)) self.subband_conv_post.apply(init_weights) self.gen_istft_n_fft = gen_istft_n_fft self.gen_istft_hop_size = gen_istft_hop_size #- for onnx if is_onnx == True: self.stft = OnnxSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) else: self.stft = TorchSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) def forward(self, x, g=None): ''' stft = TorchSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft).to(x.device) # ! ''' stft = self.stft.to(x.device) pqmf = PQMF(x.device) x = self.conv_pre(x) # [B, ch, length] for i in range(self.num_upsamples): x = F.leaky_relu(x, modules.LRELU_SLOPE) x = self.ups[i](x) xs = None for j in range(self.num_kernels): if xs is None: xs = self.resblocks[i * self.num_kernels + j](x) else: xs += self.resblocks[i * self.num_kernels + j](x) x = xs / self.num_kernels x = F.leaky_relu(x) x = self.reflection_pad(x) x = self.subband_conv_post(x) x = torch.reshape(x, (x.shape[0], self.subbands, x.shape[1] // self.subbands, x.shape[-1])) spec = torch.exp(x[:, :, :self.post_n_fft // 2 + 1, :]) phase = math.pi * torch.sin(x[:, :, self.post_n_fft // 2 + 1:, :]) y_mb_hat = stft.inverse( torch.reshape(spec, (spec.shape[0] * self.subbands, self.gen_istft_n_fft // 2 + 1, spec.shape[-1])), torch.reshape(phase, (phase.shape[0] * self.subbands, self.gen_istft_n_fft // 2 + 1, phase.shape[-1]))) y_mb_hat = torch.reshape(y_mb_hat, (x.shape[0], self.subbands, 1, y_mb_hat.shape[-1])) y_mb_hat = y_mb_hat.squeeze(-2) y_g_hat = pqmf.synthesis(y_mb_hat) return y_g_hat, y_mb_hat def remove_weight_norm(self): print('Removing weight norm...') for l in self.ups: remove_weight_norm(l) for l in self.resblocks: l.remove_weight_norm() class Multistream_iSTFT_Generator(torch.nn.Module): def __init__(self, initial_channel, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, subbands, gin_channels=0, is_onnx=False): super(Multistream_iSTFT_Generator, self).__init__() # self.h = h self.subbands = subbands self.num_kernels = len(resblock_kernel_sizes) self.num_upsamples = len(upsample_rates) self.conv_pre = weight_norm(Conv1d(initial_channel, upsample_initial_channel, 7, 1, padding=3)) resblock = modules.ResBlock1 if resblock == '1' else modules.ResBlock2 self.ups = nn.ModuleList() for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)): self.ups.append(weight_norm( ConvTranspose1d(upsample_initial_channel // (2 ** i), upsample_initial_channel // (2 ** (i + 1)), k, u, padding=(k - u) // 2))) self.resblocks = nn.ModuleList() for i in range(len(self.ups)): ch = upsample_initial_channel // (2 ** (i + 1)) for j, (k, d) in enumerate(zip(resblock_kernel_sizes, resblock_dilation_sizes)): self.resblocks.append(resblock(ch, k, d)) self.post_n_fft = gen_istft_n_fft self.ups.apply(init_weights) self.reflection_pad = torch.nn.ReflectionPad1d((1, 0)) self.reshape_pixelshuffle = [] self.subband_conv_post = weight_norm(Conv1d(ch, self.subbands * (self.post_n_fft + 2), 7, 1, padding=3)) self.subband_conv_post.apply(init_weights) self.gen_istft_n_fft = gen_istft_n_fft self.gen_istft_hop_size = gen_istft_hop_size updown_filter = torch.zeros((self.subbands, self.subbands, self.subbands)).float() for k in range(self.subbands): updown_filter[k, k, 0] = 1.0 self.register_buffer("updown_filter", updown_filter) #self.multistream_conv_post = weight_norm(Conv1d(4, 1, kernel_size=63, bias=False, padding=get_padding(63, 1))) self.multistream_conv_post = weight_norm(Conv1d(self.subbands, 1, kernel_size=63, bias=False, padding=get_padding(63, 1))) # from MB-iSTFT-VITS-44100-Ja self.multistream_conv_post.apply(init_weights) #- for onnx if is_onnx == True: self.stft = OnnxSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) else: self.stft = TorchSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft) def forward(self, x, g=None): ''' stft = TorchSTFT(filter_length=self.gen_istft_n_fft, hop_length=self.gen_istft_hop_size, win_length=self.gen_istft_n_fft).to(x.device) # ! ''' stft = self.stft.to(x.device) # pqmf = PQMF(x.device) x = self.conv_pre(x) # [B, ch, length] for i in range(self.num_upsamples): x = F.leaky_relu(x, modules.LRELU_SLOPE) x = self.ups[i](x) xs = None for j in range(self.num_kernels): if xs is None: xs = self.resblocks[i * self.num_kernels + j](x) else: xs += self.resblocks[i * self.num_kernels + j](x) x = xs / self.num_kernels x = F.leaky_relu(x) x = self.reflection_pad(x) x = self.subband_conv_post(x) x = torch.reshape(x, (x.shape[0], self.subbands, x.shape[1] // self.subbands, x.shape[-1])) spec = torch.exp(x[:, :, :self.post_n_fft // 2 + 1, :]) phase = math.pi * torch.sin(x[:, :, self.post_n_fft // 2 + 1:, :]) y_mb_hat = stft.inverse( torch.reshape(spec, (spec.shape[0] * self.subbands, self.gen_istft_n_fft // 2 + 1, spec.shape[-1])), torch.reshape(phase, (phase.shape[0] * self.subbands, self.gen_istft_n_fft // 2 + 1, phase.shape[-1]))) y_mb_hat = torch.reshape(y_mb_hat, (x.shape[0], self.subbands, 1, y_mb_hat.shape[-1])) y_mb_hat = y_mb_hat.squeeze(-2) #y_mb_hat = F.conv_transpose1d(y_mb_hat, self.updown_filter.cuda(x.device) * self.subbands, stride=self.subbands) y_mb_hat = F.conv_transpose1d(y_mb_hat, self.updown_filter.to(x.device) * self.subbands, stride=self.subbands) y_g_hat = self.multistream_conv_post(y_mb_hat) return y_g_hat, y_mb_hat def remove_weight_norm(self): print('Removing weight norm...') for l in self.ups: remove_weight_norm(l) for l in self.resblocks: l.remove_weight_norm() class DiscriminatorP(torch.nn.Module): def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False): super(DiscriminatorP, self).__init__() self.period = period self.use_spectral_norm = use_spectral_norm norm_f = weight_norm if use_spectral_norm == False else spectral_norm self.convs = nn.ModuleList([ norm_f(Conv2d(1, 32, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))), norm_f(Conv2d(32, 128, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))), norm_f(Conv2d(128, 512, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))), norm_f(Conv2d(512, 1024, (kernel_size, 1), (stride, 1), padding=(get_padding(kernel_size, 1), 0))), norm_f(Conv2d(1024, 1024, (kernel_size, 1), 1, padding=(get_padding(kernel_size, 1), 0))), ]) self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0))) def forward(self, x): fmap = [] # 1d to 2d b, c, t = x.shape if t % self.period != 0: # pad first n_pad = self.period - (t % self.period) x = F.pad(x, (0, n_pad), "reflect") t = t + n_pad x = x.view(b, c, t // self.period, self.period) for l in self.convs: x = l(x) x = F.leaky_relu(x, modules.LRELU_SLOPE) fmap.append(x) x = self.conv_post(x) fmap.append(x) x = torch.flatten(x, 1, -1) return x, fmap class DiscriminatorS(torch.nn.Module): def __init__(self, use_spectral_norm=False): super(DiscriminatorS, self).__init__() norm_f = weight_norm if use_spectral_norm == False else spectral_norm self.convs = nn.ModuleList([ norm_f(Conv1d(1, 16, 15, 1, padding=7)), norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)), norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)), norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)), norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)), norm_f(Conv1d(1024, 1024, 5, 1, padding=2)), ]) self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1)) def forward(self, x): fmap = [] for l in self.convs: x = l(x) x = F.leaky_relu(x, modules.LRELU_SLOPE) fmap.append(x) x = self.conv_post(x) fmap.append(x) x = torch.flatten(x, 1, -1) return x, fmap class MultiPeriodDiscriminator(torch.nn.Module): def __init__(self, use_spectral_norm=False): super(MultiPeriodDiscriminator, self).__init__() periods = [2, 3, 5, 7, 11] discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)] discs = discs + [DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods] self.discriminators = nn.ModuleList(discs) def forward(self, y, y_hat): y_d_rs = [] y_d_gs = [] fmap_rs = [] fmap_gs = [] for i, d in enumerate(self.discriminators): y_d_r, fmap_r = d(y) y_d_g, fmap_g = d(y_hat) y_d_rs.append(y_d_r) y_d_gs.append(y_d_g) fmap_rs.append(fmap_r) fmap_gs.append(fmap_g) return y_d_rs, y_d_gs, fmap_rs, fmap_gs class SynthesizerTrn(nn.Module): """ Synthesizer for Training """ def __init__(self, n_vocab, spec_channels, segment_size, inter_channels, hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, n_speakers=0, gin_channels=0, use_sdp=True, ms_istft_vits=False, mb_istft_vits=False, subbands=False, istft_vits=False, is_onnx=False, **kwargs): super().__init__() self.n_vocab = n_vocab self.spec_channels = spec_channels self.inter_channels = inter_channels 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.resblock = resblock self.resblock_kernel_sizes = resblock_kernel_sizes self.resblock_dilation_sizes = resblock_dilation_sizes self.upsample_rates = upsample_rates self.upsample_initial_channel = upsample_initial_channel self.upsample_kernel_sizes = upsample_kernel_sizes self.segment_size = segment_size self.n_speakers = n_speakers self.gin_channels = gin_channels self.ms_istft_vits = ms_istft_vits self.mb_istft_vits = mb_istft_vits self.istft_vits = istft_vits self.use_spk_conditioned_encoder = kwargs.get("use_spk_conditioned_encoder", False) self.use_transformer_flows = kwargs.get("use_transformer_flows", False) self.transformer_flow_type = kwargs.get("transformer_flow_type", "mono_layer_post_residual") if self.use_transformer_flows: assert self.transformer_flow_type in AVAILABLE_FLOW_TYPES, f"transformer_flow_type must be one of {AVAILABLE_FLOW_TYPES}" self.use_sdp = use_sdp # self.use_duration_discriminator = kwargs.get("use_duration_discriminator", False) self.use_noise_scaled_mas = kwargs.get("use_noise_scaled_mas", False) self.mas_noise_scale_initial = kwargs.get("mas_noise_scale_initial", 0.01) self.noise_scale_delta = kwargs.get("noise_scale_delta", 2e-6) self.current_mas_noise_scale = self.mas_noise_scale_initial if self.use_spk_conditioned_encoder and gin_channels > 0: self.enc_gin_channels = gin_channels else: self.enc_gin_channels = 0 self.enc_p = TextEncoder(n_vocab, inter_channels, hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout, gin_channels=self.enc_gin_channels) if mb_istft_vits == True: print('Multi-band iSTFT VITS2') self.dec = Multiband_iSTFT_Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, subbands, gin_channels=gin_channels, is_onnx=is_onnx) elif ms_istft_vits == True: print('Multi-stream iSTFT VITS2') self.dec = Multistream_iSTFT_Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, subbands, gin_channels=gin_channels, is_onnx=is_onnx) elif istft_vits == True: print('iSTFT-VITS2') self.dec = iSTFT_Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gen_istft_n_fft, gen_istft_hop_size, gin_channels=gin_channels, is_onnx=is_onnx) else: print('No iSTFT arguments found in json file') self.dec = Generator(inter_channels, resblock, resblock_kernel_sizes, resblock_dilation_sizes, upsample_rates, upsample_initial_channel, upsample_kernel_sizes, gin_channels=gin_channels) # vits 2 self.enc_q = PosteriorEncoder(spec_channels, inter_channels, hidden_channels, 5, 1, 16, gin_channels=gin_channels) # self.flow = ResidualCouplingBlock(inter_channels, hidden_channels, 5, 1, 4, gin_channels=gin_channels) self.flow = ResidualCouplingTransformersBlock( inter_channels, hidden_channels, 5, 1, 4, gin_channels=gin_channels, use_transformer_flows=self.use_transformer_flows, transformer_flow_type=self.transformer_flow_type ) if use_sdp: self.dp = StochasticDurationPredictor(hidden_channels, 192, 3, 0.5, 4, gin_channels=gin_channels) else: self.dp = DurationPredictor(hidden_channels, 256, 3, 0.5, gin_channels=gin_channels) if n_speakers > 1: self.emb_g = nn.Embedding(n_speakers, gin_channels) def forward(self, x, x_lengths, y, y_lengths, sid=None): # x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths) if self.n_speakers > 0: g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1] else: g = None x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths, g=g) # vits2? z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g) z_p = self.flow(z, y_mask, g=g) with torch.no_grad(): # negative cross-entropy s_p_sq_r = torch.exp(-2 * logs_p) # [b, d, t] neg_cent1 = torch.sum(-0.5 * math.log(2 * math.pi) - logs_p, [1], keepdim=True) # [b, 1, t_s] neg_cent2 = torch.matmul(-0.5 * (z_p ** 2).transpose(1, 2), s_p_sq_r) # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s] neg_cent3 = torch.matmul(z_p.transpose(1, 2), (m_p * s_p_sq_r)) # [b, t_t, d] x [b, d, t_s] = [b, t_t, t_s] neg_cent4 = torch.sum(-0.5 * (m_p ** 2) * s_p_sq_r, [1], keepdim=True) # [b, 1, t_s] neg_cent = neg_cent1 + neg_cent2 + neg_cent3 + neg_cent4 if self.use_noise_scaled_mas: epsilon = torch.std(neg_cent) * torch.randn_like(neg_cent) * self.current_mas_noise_scale neg_cent = neg_cent + epsilon attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1) attn = monotonic_align.maximum_path(neg_cent, attn_mask.squeeze(1)).unsqueeze(1).detach() w = attn.sum(2) if self.use_sdp: l_length = self.dp(x, x_mask, w, g=g) l_length = l_length / torch.sum(x_mask) logw = self.dp(x, x_mask, g=g, reverse=True, noise_scale=1.) logw_ = torch.log(w + 1e-6) * x_mask else: logw_ = torch.log(w + 1e-6) * x_mask logw = self.dp(x, x_mask, g=g) l_length = torch.sum((logw - logw_) ** 2, [1, 2]) / torch.sum(x_mask) # for averaging # expand prior m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2) logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2) z_slice, ids_slice = commons.rand_slice_segments(z, y_lengths, self.segment_size) o, o_mb = self.dec(z_slice, g=g) return o, o_mb, l_length, attn, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q), (x, logw, logw_) def infer(self, x, x_lengths, sid=None, noise_scale=1, length_scale=1, noise_scale_w=1., max_len=None): if self.n_speakers > 0: g = self.emb_g(sid).unsqueeze(-1) # [b, h, 1] else: g = None x, m_p, logs_p, x_mask = self.enc_p(x, x_lengths, g=g) if self.use_sdp: logw = self.dp(x, x_mask, g=g, reverse=True, noise_scale=noise_scale_w) else: logw = self.dp(x, x_mask, g=g) w = torch.exp(logw) * x_mask * length_scale w_ceil = torch.ceil(w) y_lengths = torch.clamp_min(torch.sum(w_ceil, [1, 2]), 1).long() y_mask = torch.unsqueeze(commons.sequence_mask(y_lengths, None), 1).to(x_mask.dtype) attn_mask = torch.unsqueeze(x_mask, 2) * torch.unsqueeze(y_mask, -1) attn = commons.generate_path(w_ceil, attn_mask) m_p = torch.matmul(attn.squeeze(1), m_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t'] logs_p = torch.matmul(attn.squeeze(1), logs_p.transpose(1, 2)).transpose(1, 2) # [b, t', t], [b, t, d] -> [b, d, t'] z_p = m_p + torch.randn_like(m_p) * torch.exp(logs_p) * noise_scale z = self.flow(z_p, y_mask, g=g, reverse=True) o, o_mb = self.dec((z * y_mask)[:, :, :max_len], g=g) return o, o_mb, attn, y_mask, (z, z_p, m_p, logs_p) #''' ## currently vits-2 is not capable of voice conversion # comment - choihkk : Assuming the use of the ResidualCouplingTransformersLayer2 module, it seems that voice conversion is possible def voice_conversion(self, y, y_lengths, sid_src, sid_tgt): assert self.n_speakers > 0, "n_speakers have to be larger than 0." g_src = self.emb_g(sid_src).unsqueeze(-1) g_tgt = self.emb_g(sid_tgt).unsqueeze(-1) z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g_src) z_p = self.flow(z, y_mask, g=g_src) z_hat = self.flow(z_p, y_mask, g=g_tgt, reverse=True) o_hat, o_hat_mb = self.dec(z_hat * y_mask, g=g_tgt) return o_hat, o_hat_mb, y_mask, (z, z_p, z_hat) #'''