import torch import torch.nn as nn import torch.nn.functional as F from models.diffusion.nn import timestep_embedding, normalization, zero_module, conv_nd, linear from models.diffusion.unet_diffusion import TimestepEmbedSequential, \ Downsample, Upsample, TimestepBlock from scripts.audio.gen.use_diffuse_tts import ceil_multiple from trainer.networks import register_model from utils.util import checkpoint, print_network def is_sequence(t): return t.dtype == torch.long class ResBlock(TimestepBlock): def __init__( self, channels, emb_channels, dropout, out_channels=None, dims=2, kernel_size=3, efficient_config=False, use_scale_shift_norm=False, ): super().__init__() self.channels = channels self.emb_channels = emb_channels self.dropout = dropout self.out_channels = out_channels or channels self.use_scale_shift_norm = use_scale_shift_norm padding = {1: 0, 3: 1, 5: 2}[kernel_size] eff_kernel = 1 if efficient_config else 3 eff_padding = 0 if efficient_config else 1 self.in_layers = nn.Sequential( normalization(channels), nn.SiLU(), conv_nd(dims, channels, self.out_channels, eff_kernel, padding=eff_padding), ) self.emb_layers = nn.Sequential( nn.SiLU(), linear( emb_channels, 2 * self.out_channels if use_scale_shift_norm else self.out_channels, ), ) self.out_layers = nn.Sequential( normalization(self.out_channels), nn.SiLU(), nn.Dropout(p=dropout), zero_module( conv_nd(dims, self.out_channels, self.out_channels, kernel_size, padding=padding) ), ) if self.out_channels == channels: self.skip_connection = nn.Identity() else: self.skip_connection = conv_nd(dims, channels, self.out_channels, eff_kernel, padding=eff_padding) def forward(self, x, emb): """ Apply the block to a Tensor, conditioned on a timestep embedding. :param x: an [N x C x ...] Tensor of features. :param emb: an [N x emb_channels] Tensor of timestep embeddings. :return: an [N x C x ...] Tensor of outputs. """ return checkpoint( self._forward, x, emb ) def _forward(self, x, emb): h = self.in_layers(x) emb_out = self.emb_layers(emb).type(h.dtype) while len(emb_out.shape) < len(h.shape): emb_out = emb_out[..., None] if self.use_scale_shift_norm: out_norm, out_rest = self.out_layers[0], self.out_layers[1:] scale, shift = torch.chunk(emb_out, 2, dim=1) h = out_norm(h) * (1 + scale) + shift h = out_rest(h) else: h = h + emb_out h = self.out_layers(h) return self.skip_connection(x) + h class StackedResidualBlock(TimestepBlock): def __init__(self, channels, emb_channels, dropout): super().__init__() self.emb_layers = nn.Sequential( nn.SiLU(), linear( emb_channels, 2 * channels, ), ) gc = channels // 4 self.initial_norm = nn.GroupNorm(num_groups=8, num_channels=channels) for i in range(5): out_channels = channels if i == 4 else gc self.add_module( f'conv{i + 1}', nn.Conv1d(channels + i * gc, out_channels, 3, 1, 1)) self.add_module(f'gn{i+1}', nn.GroupNorm(num_groups=8, num_channels=out_channels)) self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True) zero_module(self.conv5) self.drop = nn.Dropout(p=dropout) def forward(self, x, emb): return checkpoint(self.forward_, x, emb) def forward_(self, x, emb): emb_out = self.emb_layers(emb) scale, shift = torch.chunk(emb_out, 2, dim=1) x0 = self.initial_norm(x) * (1 + scale.unsqueeze(-1)) + shift.unsqueeze(-1) x1 = self.lrelu(self.gn1(self.conv1(x0))) x2 = self.lrelu(self.gn2(self.conv2(torch.cat((x, x1), 1)))) x3 = self.lrelu(self.gn3(self.conv3(torch.cat((x, x1, x2), 1)))) x4 = self.lrelu(self.gn4(self.conv4(torch.cat((x, x1, x2, x3), 1)))) x5 = self.conv5(torch.cat((x, x1, x2, x3, x4), 1)) x5 = self.drop(x5) return x5 + x class DiffusionWaveformGen(nn.Module): """ The full UNet model with residual blocks and timestep embedding. Customized to be conditioned on an aligned prior derived from a autoregressive GPT-style model. :param in_channels: channels in the input Tensor. :param in_latent_channels: channels from the input latent. :param model_channels: base channel count for the model. :param out_channels: channels in the output Tensor. :param num_res_blocks: number of residual blocks per downsample. :param dropout: the dropout probability. :param channel_mult: channel multiplier for each level of the UNet. :param conv_resample: if True, use learned convolutions for upsampling and downsampling. :param dims: determines if the signal is 1D, 2D, or 3D. :param use_scale_shift_norm: use a FiLM-like conditioning mechanism. :param resblock_updown: use residual blocks for up/downsampling. """ def __init__( self, model_channels=512, in_channels=64, in_mel_channels=256, conditioning_dim_factor=4, out_channels=128, # mean and variance dropout=0, channel_mult= (1,1.5,2), num_res_blocks=(1,1,0), token_conditioning_resolutions=(1,4), mid_resnet_depth=10, conv_resample=True, dims=1, use_fp16=False, kernel_size=3, scale_factor=2, time_embed_dim_multiplier=1, freeze_main_net=False, use_scale_shift_norm=True, # Parameters for regularization. unconditioned_percentage=.1, # This implements a mechanism similar to what is used in classifier-free training. # Parameters for super-sampling. super_sampling=False, super_sampling_max_noising_factor=.1, ): super().__init__() if super_sampling: in_channels *= 2 # In super-sampling mode, the LR input is concatenated directly onto the input. self.in_channels = in_channels self.model_channels = model_channels self.out_channels = out_channels self.dropout = dropout self.channel_mult = channel_mult self.conv_resample = conv_resample self.dims = dims self.super_sampling_enabled = super_sampling self.super_sampling_max_noising_factor = super_sampling_max_noising_factor self.unconditioned_percentage = unconditioned_percentage self.enable_fp16 = use_fp16 self.alignment_size = 2 ** (len(channel_mult)+1) self.freeze_main_net = freeze_main_net self.in_mel_channels = in_mel_channels padding = 1 if kernel_size == 3 else 2 time_embed_dim = model_channels * time_embed_dim_multiplier self.time_embed = nn.Sequential( linear(model_channels, time_embed_dim), nn.SiLU(), linear(time_embed_dim, time_embed_dim), ) conditioning_dim = model_channels * conditioning_dim_factor # Either code_converter or latent_converter is used, depending on what type of conditioning data is fed. # This model is meant to be able to be trained on both for efficiency purposes - it is far less computationally # complex to generate tokens, while generating latents will normally mean propagating through a deep autoregressive # transformer network. self.mel_converter = nn.Conv1d(in_mel_channels, conditioning_dim, 3, padding=1) self.unconditioned_embedding = nn.Parameter(torch.randn(1,conditioning_dim,1)) self.input_blocks = nn.ModuleList( [ TimestepEmbedSequential( conv_nd(dims, in_channels, model_channels, kernel_size, padding=padding) ) ] ) token_conditioning_blocks = [] self._feature_size = model_channels input_block_chans = [model_channels] ch = model_channels ds = 1 for level, (mult, num_blocks) in enumerate(zip(channel_mult, num_res_blocks)): if ds in token_conditioning_resolutions: token_conditioning_block = nn.Conv1d(conditioning_dim, ch, 1) token_conditioning_block.weight.data *= .02 self.input_blocks.append(token_conditioning_block) token_conditioning_blocks.append(token_conditioning_block) for _ in range(num_blocks): layers = [ ResBlock( ch, time_embed_dim, dropout, out_channels=int(mult * model_channels), dims=dims, kernel_size=kernel_size, use_scale_shift_norm=use_scale_shift_norm, ) ] ch = int(mult * model_channels) self.input_blocks.append(TimestepEmbedSequential(*layers)) self._feature_size += ch input_block_chans.append(ch) if level != len(channel_mult) - 1: out_ch = ch self.input_blocks.append( TimestepEmbedSequential( Downsample( ch, conv_resample, dims=dims, out_channels=out_ch, factor=scale_factor, ksize=3, pad=1 ) ) ) ch = out_ch input_block_chans.append(ch) ds *= 2 self._feature_size += ch self.middle_block = TimestepEmbedSequential(*[StackedResidualBlock(ch, time_embed_dim, dropout) for _ in range(mid_resnet_depth)]) self._feature_size += ch self.output_blocks = nn.ModuleList([]) for level, (mult, num_blocks) in list(enumerate(zip(channel_mult, num_res_blocks)))[::-1]: for i in range(num_blocks + 1): ich = input_block_chans.pop() layers = [ ResBlock( ch + ich, time_embed_dim, dropout, out_channels=int(model_channels * mult), dims=dims, kernel_size=kernel_size, use_scale_shift_norm=use_scale_shift_norm, ) ] ch = int(model_channels * mult) if level and i == num_blocks: out_ch = ch layers.append( Upsample(ch, conv_resample, dims=dims, out_channels=out_ch, factor=scale_factor) ) ds //= 2 self.output_blocks.append(TimestepEmbedSequential(*layers)) self._feature_size += ch self.out = nn.Sequential( normalization(ch), nn.SiLU(), zero_module(conv_nd(dims, model_channels, out_channels, kernel_size, padding=padding)), ) if self.freeze_main_net: mains = [self.time_embed, self.contextual_embedder, self.unconditioned_embedding, self.conditioning_timestep_integrator, self.input_blocks, self.middle_block, self.output_blocks, self.out] for m in mains: for p in m.parameters(): p.requires_grad = False p.DO_NOT_TRAIN = True def get_grad_norm_parameter_groups(self): if self.freeze_main_net: return {} groups = { 'input_blocks': list(self.input_blocks.parameters()), 'output_blocks': list(self.output_blocks.parameters()), 'middle_rrdb': list(self.middle_block.parameters()), } return groups def fix_alignment(self, x, aligned_conditioning): """ The UNet requires that the input is a certain multiple of 2, defined by the UNet depth. Enforce this by padding both and before forward propagation and removing the padding before returning. """ cm = ceil_multiple(x.shape[-1], self.alignment_size) if cm != 0: pc = (cm-x.shape[-1])/x.shape[-1] x = F.pad(x, (0,cm-x.shape[-1])) aligned_conditioning = F.pad(aligned_conditioning, (0,int(pc*aligned_conditioning.shape[-1]))) return x, aligned_conditioning def forward(self, x, timesteps, aligned_conditioning, conditioning_free=False): """ Apply the model to an input batch. :param x: an [N x C x ...] Tensor of inputs. :param timesteps: a 1-D batch of timesteps. :param aligned_conditioning: an aligned latent or sequence of tokens providing useful data about the sample to be produced. :param conditioning_free: When set, all conditioning inputs (including tokens and conditioning_input) will not be considered. :return: an [N x C x ...] Tensor of outputs. """ # Fix input size to the proper multiple of 2 so we don't get alignment errors going down and back up the U-net. orig_x_shape = x.shape[-1] x, aligned_conditioning = self.fix_alignment(x, aligned_conditioning) hs = [] time_emb = self.time_embed(timestep_embedding(timesteps, self.model_channels)) # Note: this block does not need to repeated on inference, since it is not timestep-dependent. if conditioning_free: code_emb = self.unconditioned_embedding.repeat(x.shape[0], 1, 1) else: code_emb = self.mel_converter(aligned_conditioning) time_emb = time_emb.float() h = x for k, module in enumerate(self.input_blocks): if isinstance(module, nn.Conv1d): h_tok = F.interpolate(module(code_emb), size=(h.shape[-1]), mode='nearest') h = h + h_tok else: h = module(h, time_emb) hs.append(h) h = self.middle_block(h, time_emb) for module in self.output_blocks: h = torch.cat([h, hs.pop()], dim=1) h = module(h, time_emb) out = self.out(h) # Involve probabilistic or possibly unused parameters in loss so we don't get DDP errors. extraneous_addition = 0 params = [self.unconditioned_embedding] for p in params: extraneous_addition = extraneous_addition + p.mean() out = out + extraneous_addition * 0 return out[:, :, :orig_x_shape] @register_model def register_unet_diffusion_waveform_gen3(opt_net, opt): return DiffusionWaveformGen(**opt_net['kwargs']) if __name__ == '__main__': clip = torch.randn(2, 64, 880) aligned_sequence = torch.randn(2,256,220) ts = torch.LongTensor([600, 600]) model = DiffusionWaveformGen() # Test with sequence aligned conditioning o = model(clip, ts, aligned_sequence) print_network(model)