uv2 add alignment head

codes/models/audio/tts/unified_voice2.py

@@ -238,22 +238,7 @@ class MelEncoder(nn.Module):
 class UnifiedVoice(nn.Module):
     def __init__(self, layers=8, model_dim=512, heads=8, max_text_tokens=120, max_mel_tokens=250, max_conditioning_inputs=1,
                  mel_length_compression=1024, number_text_tokens=256, number_mel_codes=8194, start_mel_token=8192,
-                 stop_mel_token=8193, start_text_token=None, checkpointing=True, types=1):
+                 stop_mel_token=8193, start_text_token=255, checkpointing=True, types=1, only_alignment_head=False):
-        """
-        Args:
-            layers: Number of layers in transformer stack.
-            model_dim: Operating dimensions of the transformer
-            heads: Number of transformer heads. Must be divisible by model_dim. Recommend model_dim//64
-            max_text_tokens: Maximum number of text tokens that will be encountered by model.
-            max_mel_tokens: Maximum number of MEL tokens that will be encountered by model.
-            max_conditioning_inputs: Maximum number of conditioning inputs provided to the model. If (1), conditioning input can be of format (b,80,s), otherwise (b,n,80,s).
-            mel_length_compression: The factor between <number_input_samples> and <mel_tokens>. Used to compute MEL code padding given wav input length.
-            number_text_tokens:
-            number_mel_codes:
-            start_mel_token:
-            stop_mel_token:
-            checkpointing:
-        """
         super().__init__()
 
         self.number_text_tokens = number_text_tokens
@@ -278,6 +263,15 @@ class UnifiedVoice(nn.Module):
         self.final_norm = nn.LayerNorm(model_dim)
         self.text_head = nn.Linear(model_dim, self.number_text_tokens*types+1)
         self.mel_head = nn.Linear(model_dim, self.number_mel_codes)
+        self.alignment_head = nn.Linear(model_dim, 256)
+
+        if only_alignment_head:
+            for p in self.parameters():
+                p.DO_NOT_TRAIN = True
+                p.requires_grad = False
+            for p in self.alignment_head.parameters():
+                del p.DO_NOT_TRAIN
+                p.requires_grad = True
 
         # Initialize the embeddings per the GPT-2 scheme
         embeddings = [self.text_embedding, self.mel_embedding]
@@ -310,11 +304,8 @@ class UnifiedVoice(nn.Module):
                 mel_input_tokens[b, actual_end:] = self.stop_mel_token
         return mel_input_tokens
 
-    def get_logits(self, speech_conditioning_inputs, first_inputs, first_head, second_inputs=None, second_head=None, return_latent=False):
+    def get_logits(self, speech_conditioning_inputs, first_inputs, second_inputs, return_latent=False):
-        if second_inputs is not None:
         emb = torch.cat([speech_conditioning_inputs, first_inputs, second_inputs], dim=1)
-        else:
-            emb = torch.cat([speech_conditioning_inputs, first_inputs], dim=1)
 
         gpt_out = self.gpt(inputs_embeds=emb, return_dict=True)
 
@@ -324,16 +315,19 @@ class UnifiedVoice(nn.Module):
         if return_latent:
             return enc[:, speech_conditioning_inputs.shape[1]:speech_conditioning_inputs.shape[1]+first_inputs.shape[1]], enc[:, -second_inputs.shape[1]:]
 
-        first_logits = enc[:, :first_inputs.shape[1]]
+        text_logits = enc[:, :first_inputs.shape[1]]
-        first_logits = first_head(first_logits)
+        text_logits = self.text_head(text_logits)
-        first_logits = first_logits.permute(0,2,1)
+        text_logits = text_logits.permute(0,2,1)
-        if second_inputs is not None:
+
-            second_logits = enc[:, -second_inputs.shape[1]:]
+        mel_logits = enc[:, -second_inputs.shape[1]:]
-            second_logits = second_head(second_logits)
+        mel_logits = self.mel_head(mel_logits)
-            second_logits = second_logits.permute(0,2,1)
+        mel_logits = mel_logits.permute(0,2,1)
-            return first_logits, second_logits
+
-        else:
+        alignment_logits = enc[:, -second_inputs.shape[1]:]
-            return first_logits
+        alignment_logits = self.alignment_head(alignment_logits)
+        alignment_logits = alignment_logits.permute(0,2,1)
+
+        return text_logits, mel_logits, alignment_logits
 
 
     def get_conditioning_latent(self, speech_conditioning_input):
@@ -346,7 +340,7 @@ class UnifiedVoice(nn.Module):
         return conds
 
 
-    def forward(self, speech_conditioning_input, text_inputs, text_lengths, mel_codes, wav_lengths, types=None, text_first=True, return_latent=False):
+    def forward(self, speech_conditioning_input, text_inputs, text_lengths, mel_codes, ctc_codes, wav_lengths, types=None, return_latent=False):
         """
         Forward pass that uses both text and voice in either text conditioning mode or voice conditioning mode
         (actuated by `text_first`).
@@ -363,12 +357,22 @@ class UnifiedVoice(nn.Module):
         if types is not None:
             text_inputs = text_inputs * (1+types).unsqueeze(-1)
 
+        # TODO: do this in the dataloader.
+        for b in range(ctc_codes.shape[0]):
+            last_code = 0
+            for j in range(ctc_codes.shape[1]):
+                if ctc_codes[b][j] == 0:
+                    ctc_codes[b][j] = last_code
+                else:
+                    last_code = ctc_codes[b][j]
+        alignment_targets = F.interpolate(ctc_codes.unsqueeze(1).float(), size=(mel_codes.shape[-1],), mode='nearest').long().squeeze()
+
         mel_codes = self.set_mel_padding(mel_codes, wav_lengths)
         text_inputs = F.pad(text_inputs, (0,1), value=self.stop_text_token)
         mel_codes = F.pad(mel_codes, (0,1), value=self.stop_mel_token)
+        alignment_targets = F.pad(alignment_targets, (0,2), value=0)
 
         conds = self.get_conditioning_latent(speech_conditioning_input)
-
         text_inputs, text_targets = self.build_aligned_inputs_and_targets(text_inputs, self.start_text_token, self.stop_text_token)
         text_emb = self.text_embedding(text_inputs) + self.text_pos_embedding(text_inputs)
         mel_codes, mel_targets = self.build_aligned_inputs_and_targets(mel_codes, self.start_mel_token, self.stop_mel_token)
@@ -376,18 +380,14 @@ class UnifiedVoice(nn.Module):
         mel_emb = self.mel_embedding(mel_inp)
         mel_emb = mel_emb + self.mel_pos_embedding(mel_codes)
 
-        if text_first:
+        text_logits, mel_logits, alignment_logits = self.get_logits(conds, text_emb, mel_emb, return_latent=return_latent)
-            text_logits, mel_logits = self.get_logits(conds, text_emb, self.text_head, mel_emb, self.mel_head, return_latent=return_latent)
         if return_latent:
             return mel_logits[:, :-2]  # Despite the name, these are not logits. Strip off the two tokens added by this forward pass.
-        else:
-            mel_logits, text_logits = self.get_logits(conds, mel_emb, self.mel_head, text_emb, self.text_head, return_latent=return_latent)
-            if return_latent:
-                return text_logits[:, :-2]  # Despite the name, these are not logits. Strip off the two tokens added by this forward pass.
 
         loss_text = F.cross_entropy(text_logits, text_targets.long())
         loss_mel = F.cross_entropy(mel_logits, mel_targets.long())
-        return loss_text.mean(), loss_mel.mean(), mel_logits
+        loss_alignment = F.cross_entropy(alignment_logits, alignment_targets)
+        return loss_text.mean(), loss_mel.mean(), loss_alignment, mel_logits
 
     def inference_speech(self, speech_conditioning_input, text_inputs, **hf_generate_kwargs):
         if self.max_mel_tokens == -1:  # Assume if this is the case, max_mel_tokens=-1 also