dataset improvements and fix to unified_voice_Bilevel

codes/data/audio/paired_voice_audio_dataset.py

@@ -135,8 +135,8 @@ class TextWavLoader(torch.utils.data.Dataset):
             (self.max_text_len is not None and tseq.shape[0] > self.max_text_len):
             # Basically, this audio file is nonexistent or too long to be supported by the dataset.
             # It's hard to handle this situation properly. Best bet is to return the a random valid token and skew the dataset somewhat as a result.
-            #if wav is not None:
+            if self.debug_failures:
-            #    print(f"Exception {index} wav_len:{wav.shape[-1]} text_len:{tseq.shape[0]} fname: {path}")
+                print(f"error loading {path}: ranges are out of bounds; {wav.shape[-1]}, {tseq.shape[0]}")
             rv = random.randint(0,len(self)-1)
             return self[rv]
         orig_output = wav.shape[-1]

codes/data/audio/unsupervised_audio_dataset.py

@@ -85,7 +85,10 @@ def load_similar_clips(path, sample_length, sample_rate, n=3, include_self=True,
             rand_start = random.randint(0, gap)
             rel_clip = rel_clip[:, rand_start:rand_start+sample_length]
         related_clips.append(rel_clip)
-    return torch.stack(related_clips, dim=0)
+    if n > 1:
+        return torch.stack(related_clips, dim=0)
+    else:
+        return related_clips[0]
 
 
 class UnsupervisedAudioDataset(torch.utils.data.Dataset):

codes/models/gpt_voice/unified_voice_bilevel.py

@@ -1,4 +1,5 @@
 import functools
+from math import log
 
 import torch
 import torch.nn as nn
@@ -12,6 +13,10 @@ from trainer.networks import register_model
 from utils.util import opt_get
 
 
+def null_position_embeddings(range, dim):
+    return torch.zeros((range.shape[0], range.shape[1], dim), device=range.device)
+
+
 class ConditioningEncoder(nn.Module):
     def __init__(self,
                  spec_dim,
@@ -34,8 +39,22 @@ class ConditioningEncoder(nn.Module):
         return h[:, :, 0]
 
 
-def null_position_embeddings(range, dim):
+class TopEncoder(nn.Module):
-    return torch.zeros((range.shape[0], range.shape[1], dim), device=range.device)
+    def __init__(self, layers, dim, heads, do_checkpointing=False, dim_reduction=16):
+        self.init = nn.Conv1d(dim, dim, kernel_size=1)
+        reduction_layers = []
+        for j in range(int(log(dim_reduction, 2))):
+            reduction_layers.append(AttentionBlock(dim, heads, do_checkpoint=do_checkpointing))
+            reduction_layers.append(nn.Conv1d(dim, dim, kernel_size=3, padding=1, stride=2))
+        self.reduction_layers = nn.Sequential(*reduction_layers)
+        actual_layers = [AttentionBlock(dim, heads, do_checkpoint=do_checkpointing) for _ in range(layers)]
+        self.actual_layers = nn.Sequential(*actual_layers)
+
+    def forward(self, x):
+        h = self.init(x)
+        h = self.reduction_layers(h)
+        h = self.actual_layers(h)
+        return h
 
 
 class UnifiedGptVoice(nn.Module):
@@ -47,7 +66,8 @@ class UnifiedGptVoice(nn.Module):
     - Voice conditioned on text
     """
 
-    def __init__(self, layers=8, model_dim=512, heads=8, max_symbols_per_phrase=120, max_mel_tokens=250, max_total_tokens=370, max_conditioning_inputs=3,
+    def __init__(self, top_encoder_layers=4, top_layers=8, bottom_layers=8, top_dim_reduction=16, model_dim=512, heads=8,
+                 max_symbols_per_phrase=120, max_mel_tokens=250, max_total_tokens=370, max_conditioning_inputs=3,
                  checkpointing=True, mel_length_compression=1024, max_conditioning_length=60, number_text_tokens=256,
                  start_text_token=255, stop_text_token=0, number_mel_codes=8194, start_mel_token=8192,
                  stop_mel_token=8193):
@@ -73,18 +93,35 @@ class UnifiedGptVoice(nn.Module):
         self.mel_pos_solo_embedding = nn.Embedding(self.max_mel_tokens + 1, model_dim)
         self.mel_pos_paired_embedding = nn.Embedding(self.max_mel_tokens + 1, model_dim)
         seq_length = 2+self.max_total_tokens+self.max_conditioning_inputs
-        self.gpt_config = GPT2Config(vocab_size=self.number_mel_codes,
+
-                                     n_positions=seq_length,
+        self.top_encoder = TopEncoder(top_encoder_layers, model_dim, heads, do_checkpointing=checkpointing,
-                                     n_ctx=seq_length,
+                                      dim_reduction=top_dim_reduction)
-                                     n_embd=model_dim,
+        self.top_gpt_config = GPT2Config(vocab_size=1,
-                                     n_layer=layers,
+                                         n_positions=seq_length // top_dim_reduction,
-                                     n_head=heads,
+                                         n_ctx=seq_length // top_dim_reduction,
-                                     gradient_checkpointing=checkpointing,
+                                         n_embd=model_dim,
-                                     use_cache=not checkpointing)
+                                         n_layer=top_layers,
-        self.gpt = GPT2Model(self.gpt_config)
+                                         n_head=heads,
+                                         gradient_checkpointing=checkpointing,
+                                         use_cache=not checkpointing)
+        self.top_gpt = GPT2Model(self.top_gpt_config)
+        del self.top_gpt.wte
+        self.top_gpt_start_embedding = nn.Parameter(torch.randn(1,1,model_dim)*self.top_gpt_config.initializer_range,
+                                                    requires_grad=True)
+        self.top_dim_reduction = top_dim_reduction
+
+        self.bottom_gpt_config = GPT2Config(vocab_size=self.number_mel_codes,
+                                            n_positions=seq_length,
+                                            n_ctx=seq_length,
+                                            n_embd=model_dim,
+                                            n_layer=bottom_layers,
+                                            n_head=heads,
+                                            gradient_checkpointing=checkpointing,
+                                            use_cache=not checkpointing)
+        self.bottom_gpt = GPT2Model(self.bottom_gpt_config)
         # Override the built in positional embeddings
-        del self.gpt.wpe
+        del self.bottom_gpt.wpe
-        self.gpt.wpe = functools.partial(null_position_embeddings, dim=model_dim)
+        self.bottom_gpt.wpe = functools.partial(null_position_embeddings, dim=model_dim)
 
         self.final_norm = nn.LayerNorm(model_dim)
         self.text_head = nn.Linear(model_dim, self.number_text_tokens)
@@ -94,7 +131,7 @@ class UnifiedGptVoice(nn.Module):
         # Initialize the embeddings per the GPT-2 scheme
         for module in [self.text_embedding, self.text_pos_solo_embedding, self.text_pos_paired_embedding,
                        self.mel_pos_solo_embedding, self.mel_pos_paired_embedding]:
-            module.weight.data.normal_(mean=0.0, std=self.gpt.config.initializer_range)
+            module.weight.data.normal_(mean=0.0, std=self.bottom_gpt.config.initializer_range)
             if module.padding_idx is not None:
                 module.weight.data[module.padding_idx].zero_()
 
@@ -129,13 +166,34 @@ class UnifiedGptVoice(nn.Module):
             cond_input = cond_input[:,:,:self.max_conditioning_length]
         return cond_input
 
+
+    def get_top_embeddings(self, embedded_input):
+        true_embeddings = self.top_encoder(embedded_input)
+        inputs = torch.cat([self.top_gpt_start_embedding, true_embeddings[:,:-1]], dim=1)
+        top_pred = self.top_gpt(inputs_embeds=inputs, return_dict=True)
+        return top_pred.last_hidden_state, true_embeddings
+
+
+    def inject_top_embeddings(self, embedded_input, probability_of_true_top_embedding=.5):
+        pred, true = self.get_top_embeddings(embedded_input)
+        rand = torch.bernoulli(torch.full((1,embedded_input.shape[1]),
+                                          fill_value=probability_of_true_top_embedding)).to(embedded_input.device)
+        mix = pred * rand + true * (not rand)
+        embs = torch.chunk(embedded_input, self.top_dim_reduction, dim=1)
+        assert len(embs) == mix.shape[1]
+        rejoin = []
+        for i, emb in enumerate(embs):
+            rejoin.append(torch.cat([mix[i], emb]), dim=1)
+        return torch.cat(rejoin, dim=1)
+
+
     def get_logits(self, speech_conditioning_input, first_inputs, first_head, second_inputs=None, second_head=None, get_attns=False):
         if second_inputs is not None:
             emb = torch.cat([speech_conditioning_input, first_inputs, second_inputs], dim=1)
         else:
             emb = torch.cat([speech_conditioning_input, first_inputs], dim=1)
 
-        gpt_out = self.gpt(inputs_embeds=emb, return_dict=True, output_attentions=get_attns)
+        gpt_out = self.bottom_gpt(inputs_embeds=emb, return_dict=True, output_attentions=get_attns)
         if get_attns:
             return gpt_out.attentions
 
@@ -173,8 +231,9 @@ class UnifiedGptVoice(nn.Module):
         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_paired_embedding(torch.arange(text_inputs.shape[1], device=text_inputs.device))
         mel_inputs, mel_targets = self.build_aligned_inputs_and_targets(mel_inputs, self.start_mel_token, self.stop_mel_token)
-        mel_emb = self.gpt.get_input_embeddings()(mel_inputs)
+        mel_emb = self.bottom_gpt.get_input_embeddings()(mel_inputs)
         mel_emb = mel_emb + self.mel_pos_paired_embedding(torch.arange(mel_emb.shape[1], device=mel_emb.device))
+
         if text_first:
             text_logits, mel_logits = self.get_logits(speech_conditioning_input, text_emb, self.text_head, mel_emb, self.mel_head, get_attns=return_attentions)
         else:
@@ -213,7 +272,7 @@ class UnifiedGptVoice(nn.Module):
         speech_conditioning_input = self.conditioning_encoder(speech_conditioning_input).unsqueeze(1)
 
         mel_inputs, mel_targets = self.build_aligned_inputs_and_targets(mel_inputs, self.start_mel_token, self.stop_mel_token)
-        mel_emb = self.gpt.get_input_embeddings()(mel_inputs)
+        mel_emb = self.bottom_gpt.get_input_embeddings()(mel_inputs)
         mel_emb = mel_emb + self.mel_pos_solo_embedding(torch.arange(mel_emb.shape[1], device=mel_emb.device))
         mel_logits = self.get_logits(speech_conditioning_input, mel_emb, self.mel_head)
         loss_mel = F.cross_entropy(mel_logits, mel_targets.long())
@@ -221,7 +280,7 @@ class UnifiedGptVoice(nn.Module):
 
     def inference_speech(self, speech_conditioning_input, text_inputs, **hf_generate_kwargs):
         if not hasattr(self, 'inference_model'):
-            self.inference_model = GPT2InferenceModel(self.gpt_config, self.gpt, self.mel_pos_paired_embedding, self.final_norm, self.mel_head)
+            self.inference_model = GPT2InferenceModel(self.bottom_gpt_config, self.bottom_gpt, self.mel_pos_paired_embedding, self.final_norm, self.mel_head)
 
         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_paired_embedding(torch.arange(text_inputs.shape[1], device=text_inputs.device))
@@ -237,12 +296,12 @@ class UnifiedGptVoice(nn.Module):
         fake_inputs[:,-1] = self.start_mel_token
 
         gen = self.inference_model.generate(fake_inputs, bos_token_id=self.start_mel_token, pad_token_id=self.stop_mel_token, eos_token_id=self.stop_mel_token,
-                                            max_length=self.gpt_config.n_positions, **hf_generate_kwargs)
+                                            max_length=self.bottom_gpt_config.n_positions, **hf_generate_kwargs)
         return gen[:, fake_inputs.shape[1]:]
 
 
 @register_model
-def register_unified_gpt_voice(opt_net, opt):
+def register_unified_gpt_voice_bilevel(opt_net, opt):
     return UnifiedGptVoice(**opt_get(opt_net, ['kwargs'], {}))