import torch import torch.nn as nn import torch.nn.functional as F from einops import rearrange from torch import einsum from models.lucidrains.dalle.transformer import Transformer from trainer.networks import register_model from utils.util import opt_get def exists(val): return val is not None def masked_mean(t, mask, dim = 1): t = t.masked_fill(~mask[:, :, None], 0.) return t.sum(dim = 1) / mask.sum(dim = 1)[..., None] class MelTextCLIP(nn.Module): """ CLIP model retrofitted for performing contrastive evaluation between MEL data and the corresponding transcribed text. Originally from https://github.com/lucidrains/DALLE-pytorch/blob/main/dalle_pytorch/dalle_pytorch.py """ def __init__( self, *, dim_text=512, dim_speech=512, dim_latent=512, num_text_tokens=256, text_enc_depth=6, text_seq_len=120, text_heads=8, num_speech_tokens=8192, speech_enc_depth=6, speech_heads=8, speech_seq_len=250, text_mask_percentage=0, voice_mask_percentage=0, mel_compression=256, ): super().__init__() self.text_emb = nn.Embedding(num_text_tokens, dim_text) self.text_pos_emb = nn.Embedding(text_seq_len, dim_text) self.text_transformer = Transformer(causal=False, seq_len=text_seq_len, dim=dim_text, depth=text_enc_depth, heads=text_heads, rotary_emb=False) self.to_text_latent = nn.Linear(dim_text, dim_latent, bias=False) self.speech_enc = nn.Conv1d(80, dim_speech, kernel_size=3, padding=1) self.speech_pos_emb = nn.Embedding(num_speech_tokens, dim_speech) self.speech_transformer = Transformer(causal=False, seq_len=speech_seq_len, dim=dim_speech, depth=speech_enc_depth, heads=speech_heads, rotary_emb=False) self.to_speech_latent = nn.Linear(dim_speech, dim_latent, bias=False) self.temperature = nn.Parameter(torch.tensor(1.)) self.text_mask_percentage = text_mask_percentage self.voice_mask_percentage = voice_mask_percentage self.mel_compression = mel_compression def get_text_projections(self, text, text_mask=None): if text_mask is None: text_mask = torch.ones_like(text.float()).bool() text_emb = self.text_emb(text) text_emb += self.text_pos_emb(torch.arange(text.shape[1], device=text.device)) with torch.autocast(text.device.type): enc_text = self.text_transformer(text_emb, mask=text_mask) text_latents = masked_mean(enc_text, text_mask, dim=1) return self.to_text_latent(text_latents).float() def get_speech_projection(self, mel, voice_mask=None): if voice_mask is None: voice_mask = torch.ones_like(mel[:,0,:].float()).bool() speech_emb = self.speech_enc(mel).permute(0,2,1) speech_emb += self.speech_pos_emb(torch.arange(speech_emb.shape[1], device=mel.device)) with torch.autocast(speech_emb.device.type): enc_speech = self.speech_transformer(speech_emb, mask=voice_mask) speech_latents = masked_mean(enc_speech, voice_mask, dim=1) return self.to_speech_latent(speech_latents).float() def forward( self, text, text_lengths, mel, wav_lengths, return_loss=False ): # This model will receive micro-batches with a ton of padding for both the text and MELs. Ameliorate this by # chopping the inputs by the maximum actual length. max_text_len = text_lengths.max() text = text[:, :max_text_len] max_mel_len = wav_lengths.max() // self.mel_compression mel = mel[:, :, :max_mel_len] b, device = text.shape[0], text.device if self.training: text_mask = torch.rand_like(text.float()) > self.text_mask_percentage voice_mask = torch.rand_like(mel[:,0,:].float()) > self.voice_mask_percentage else: text_mask = torch.ones_like(text.float()).bool() voice_mask = torch.ones_like(mel[:,0,:].float()).bool() text_emb = self.text_emb(text) text_emb += self.text_pos_emb(torch.arange(text.shape[1], device=device)) speech_emb = self.speech_enc(mel).permute(0,2,1) speech_emb += self.speech_pos_emb(torch.arange(speech_emb.shape[1], device=device)) # Only autocast the transformer part. The MEL encoder loses accuracy if you autcast it. with torch.autocast(speech_emb.device.type): enc_text = self.text_transformer(text_emb, mask=text_mask) enc_speech = self.speech_transformer(speech_emb, mask=voice_mask) text_latents = masked_mean(enc_text, text_mask, dim=1) speech_latents = masked_mean(enc_speech, voice_mask, dim=1) text_latents = self.to_text_latent(text_latents).float() speech_latents = self.to_speech_latent(speech_latents).float() text_latents, speech_latents = map(lambda t: F.normalize(t, p=2, dim=-1), (text_latents, speech_latents)) temp = self.temperature.exp() if not return_loss: sim = einsum('n d, n d -> n', text_latents, speech_latents) * temp return sim sim = einsum('i d, j d -> i j', text_latents, speech_latents) * temp labels = torch.arange(b, device=device) loss = (F.cross_entropy(sim, labels) + F.cross_entropy(sim.t(), labels)) / 2 return loss @register_model def register_mel_text_clip(opt_net, opt): return MelTextCLIP(**opt_get(opt_net, ['kwargs'], {})) if __name__ == '__main__': clip = MelTextCLIP(text_mask_percentage=.2, voice_mask_percentage=.2) clip(torch.randint(0,256,(2,120)), torch.tensor([50,100]), torch.randn(2,80,400), torch.tensor([10100,10200]), return_loss=True)