from collections import deque from functools import partial from inspect import isfunction import numpy as np import torch import torch.nn.functional as F from torch import nn from tqdm import tqdm from modules.encoder import SvcEncoder from training.train_pipeline import Batch2Loss from utils.hparams import hparams def exists(x): return x is not None def default(val, d): if exists(val): return val return d() if isfunction(d) else d # gaussian diffusion trainer class def extract(a, t, x_shape): b, *_ = t.shape out = a.gather(-1, t) return out.reshape(b, *((1,) * (len(x_shape) - 1))) def noise_like(shape, device, repeat=False): repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1))) noise = lambda: torch.randn(shape, device=device) return repeat_noise() if repeat else noise() def linear_beta_schedule(timesteps, max_beta=hparams.get('max_beta', 0.01)): """ linear schedule """ betas = np.linspace(1e-4, max_beta, timesteps) return betas def cosine_beta_schedule(timesteps, s=0.008): """ cosine schedule as proposed in https://openreview.net/forum?id=-NEXDKk8gZ """ steps = timesteps + 1 x = np.linspace(0, steps, steps) alphas_cumprod = np.cos(((x / steps) + s) / (1 + s) * np.pi * 0.5) ** 2 alphas_cumprod = alphas_cumprod / alphas_cumprod[0] betas = 1 - (alphas_cumprod[1:] / alphas_cumprod[:-1]) return np.clip(betas, a_min=0, a_max=0.999) beta_schedule = { "cosine": cosine_beta_schedule, "linear": linear_beta_schedule, } class GaussianDiffusion(nn.Module): def __init__(self, phone_encoder, out_dims, denoise_fn, timesteps=1000, K_step=1000, loss_type=hparams.get('diff_loss_type', 'l1'), betas=None, spec_min=None, spec_max=None): super().__init__() self.denoise_fn = denoise_fn self.fs2 = SvcEncoder(phone_encoder, out_dims) self.mel_bins = out_dims if exists(betas): betas = betas.detach().cpu().numpy() if isinstance(betas, torch.Tensor) else betas else: if 'schedule_type' in hparams.keys(): betas = beta_schedule[hparams['schedule_type']](timesteps) else: betas = cosine_beta_schedule(timesteps) alphas = 1. - betas alphas_cumprod = np.cumprod(alphas, axis=0) alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1]) timesteps, = betas.shape self.num_timesteps = int(timesteps) self.K_step = K_step self.loss_type = loss_type self.noise_list = deque(maxlen=4) to_torch = partial(torch.tensor, dtype=torch.float32) self.register_buffer('betas', to_torch(betas)) self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod)) self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev)) # calculations for diffusion q(x_t | x_{t-1}) and others self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod))) self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod))) self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod))) self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod))) self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1))) # calculations for posterior q(x_{t-1} | x_t, x_0) posterior_variance = betas * (1. - alphas_cumprod_prev) / (1. - alphas_cumprod) # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t) self.register_buffer('posterior_variance', to_torch(posterior_variance)) # below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain self.register_buffer('posterior_log_variance_clipped', to_torch(np.log(np.maximum(posterior_variance, 1e-20)))) self.register_buffer('posterior_mean_coef1', to_torch( betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))) self.register_buffer('posterior_mean_coef2', to_torch( (1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod))) self.register_buffer('spec_min', torch.FloatTensor(spec_min)[None, None, :hparams['keep_bins']]) self.register_buffer('spec_max', torch.FloatTensor(spec_max)[None, None, :hparams['keep_bins']]) def predict_start_from_noise(self, x_t, t, noise): return ( extract(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - extract(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise ) def q_posterior(self, x_start, x_t, t): posterior_mean = ( extract(self.posterior_mean_coef1, t, x_t.shape) * x_start + extract(self.posterior_mean_coef2, t, x_t.shape) * x_t ) posterior_variance = extract(self.posterior_variance, t, x_t.shape) posterior_log_variance_clipped = extract(self.posterior_log_variance_clipped, t, x_t.shape) return posterior_mean, posterior_variance, posterior_log_variance_clipped def p_mean_variance(self, x, t, cond, clip_denoised: bool): noise_pred = self.denoise_fn(x, t, cond=cond) x_recon = self.predict_start_from_noise(x, t=t, noise=noise_pred) if clip_denoised: x_recon.clamp_(-1., 1.) model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t) return model_mean, posterior_variance, posterior_log_variance @torch.no_grad() def p_sample(self, x, t, cond, clip_denoised=True, repeat_noise=False): b, *_, device = *x.shape, x.device model_mean, _, model_log_variance = self.p_mean_variance(x=x, t=t, cond=cond, clip_denoised=clip_denoised) noise = noise_like(x.shape, device, repeat_noise) # no noise when t == 0 nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1))) return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise @torch.no_grad() def p_sample_plms(self, x, t, interval, cond): """ Use the PLMS method from [Pseudo Numerical Methods for Diffusion Models on Manifolds](https://arxiv.org/abs/2202.09778). """ def get_x_pred(x, noise_t, t): a_t = extract(self.alphas_cumprod, t, x.shape) a_prev = extract(self.alphas_cumprod, torch.max(t - interval, torch.zeros_like(t)), x.shape) a_t_sq, a_prev_sq = a_t.sqrt(), a_prev.sqrt() x_delta = (a_prev - a_t) * ((1 / (a_t_sq * (a_t_sq + a_prev_sq))) * x - 1 / ( a_t_sq * (((1 - a_prev) * a_t).sqrt() + ((1 - a_t) * a_prev).sqrt())) * noise_t) x_pred = x + x_delta return x_pred noise_list = self.noise_list noise_pred = self.denoise_fn(x, t, cond=cond) if len(noise_list) == 0: x_pred = get_x_pred(x, noise_pred, t) noise_pred_prev = self.denoise_fn(x_pred, max(t - interval, 0), cond=cond) noise_pred_prime = (noise_pred + noise_pred_prev) / 2 elif len(noise_list) == 1: noise_pred_prime = (3 * noise_pred - noise_list[-1]) / 2 elif len(noise_list) == 2: noise_pred_prime = (23 * noise_pred - 16 * noise_list[-1] + 5 * noise_list[-2]) / 12 elif len(noise_list) >= 3: noise_pred_prime = (55 * noise_pred - 59 * noise_list[-1] + 37 * noise_list[-2] - 9 * noise_list[-3]) / 24 x_prev = get_x_pred(x, noise_pred_prime, t) noise_list.append(noise_pred) return x_prev def q_sample(self, x_start, t, noise=None): noise = default(noise, lambda: torch.randn_like(x_start)) return ( extract(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start + extract(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise ) def p_losses(self, x_start, t, cond, noise=None, nonpadding=None): noise = default(noise, lambda: torch.randn_like(x_start)) x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) x_recon = self.denoise_fn(x_noisy, t, cond) if self.loss_type == 'l1': if nonpadding is not None: loss = ((noise - x_recon).abs() * nonpadding.unsqueeze(1)).mean() else: # print('are you sure w/o nonpadding?') loss = (noise - x_recon).abs().mean() elif self.loss_type == 'l2': loss = F.mse_loss(noise, x_recon) else: raise NotImplementedError() return loss def forward(self, hubert, mel2ph=None, spk_embed_id=None, ref_mels=None, f0=None, energy=None, infer=False): ''' conditioning diffusion, use fastspeech2 encoder output as the condition ''' ret = self.fs2(hubert, mel2ph, spk_embed_id, f0, energy) cond = ret['decoder_inp'].transpose(1, 2) b, *_, device = *hubert.shape, hubert.device if not infer: Batch2Loss.module4( self.p_losses, self.norm_spec(ref_mels), cond, ret, self.K_step, b, device ) else: t = self.K_step shape = (cond.shape[0], 1, self.mel_bins, cond.shape[2]) x = torch.randn(shape, device=device) if hparams.get('pndm_speedup') and hparams['pndm_speedup'] > 1: self.noise_list = deque(maxlen=4) iteration_interval = hparams['pndm_speedup'] for i in tqdm(reversed(range(0, t, iteration_interval)), desc='sample time step', total=t // iteration_interval): x = self.p_sample_plms(x, torch.full((b,), i, device=device, dtype=torch.long), iteration_interval, cond) else: for i in tqdm(reversed(range(0, t)), desc='sample time step', total=t): x = self.p_sample(x, torch.full((b,), i, device=device, dtype=torch.long), cond) x = x[:, 0].transpose(1, 2) if mel2ph is not None: # for singing ret['mel_out'] = self.denorm_spec(x) * ((mel2ph > 0).float()[:, :, None]) else: ret['mel_out'] = self.denorm_spec(x) return ret def norm_spec(self, x): return (x - self.spec_min) / (self.spec_max - self.spec_min) * 2 - 1 def denorm_spec(self, x): return (x + 1) / 2 * (self.spec_max - self.spec_min) + self.spec_min