Merge pull request #499 from WatchTower-Liu/inference_optin

新增无参考文本模式

.gitignore

@@ -7,5 +7,8 @@ runtime
 output
 logs
 reference
+GPT_weights
 SoVITS_weights
-GPT_weights
+TEMP
+
+

GPT_SoVITS/AR/models/t2s_model.py

@@ -337,7 +337,7 @@ class Text2SemanticDecoder(nn.Module):
 
         # AR Decoder
         y = prompts
-        prefix_len = y.shape[1]
+        
         x_len = x.shape[1]
         x_attn_mask = torch.zeros((x_len, x_len), dtype=torch.bool)
         stop = False
@@ -353,47 +353,41 @@ class Text2SemanticDecoder(nn.Module):
             "first_infer": 1,
             "stage": 0,
         }
-        for idx in tqdm(range(1500)):
+        ###################  first step ##########################
-            if cache["first_infer"] == 1:
+        if y is not None:
-                y_emb = self.ar_audio_embedding(y)
+            y_emb = self.ar_audio_embedding(y)
-            else:
+            y_len = y_emb.shape[1]
-                y_emb = torch.cat(
+            prefix_len = y.shape[1]
-                    [cache["y_emb"], self.ar_audio_embedding(y[:, -1:])], 1
-                )
-            cache["y_emb"] = y_emb
             y_pos = self.ar_audio_position(y_emb)
-            # x 和逐渐增长的 y 一起输入给模型
+            xy_pos = torch.concat([x, y_pos], dim=1)
-            if cache["first_infer"] == 1:
+            cache["y_emb"] = y_emb
-                xy_pos = torch.concat([x, y_pos], dim=1)
+            ref_free = False
-            else:
+        else:
-                xy_pos = y_pos[:, -1:]
+            y_emb = None
-            y_len = y_pos.shape[1]
+            y_len = 0
-            ###以下3个不做缓存
+            prefix_len = 0
-            if cache["first_infer"] == 1:
+            y_pos = None
-                x_attn_mask_pad = F.pad(
+            xy_pos = x
+            y = torch.zeros(x.shape[0], 0, dtype=torch.int, device=x.device)
+            ref_free = True
+
+        x_attn_mask_pad = F.pad(
                     x_attn_mask,
                     (0, y_len),  ###xx的纯0扩展到xx纯0+xy纯1，(x,x+y)
                     value=True,
                 )
-                y_attn_mask = F.pad(  ###yy的右上1扩展到左边xy的0,(y,x+y)
+        y_attn_mask = F.pad(  ###yy的右上1扩展到左边xy的0,(y,x+y)
-                    torch.triu(torch.ones(y_len, y_len, dtype=torch.bool), diagonal=1),
+            torch.triu(torch.ones(y_len, y_len, dtype=torch.bool), diagonal=1),
-                    (x_len, 0),
+            (x_len, 0),
-                    value=False,
+            value=False,
-                )
+        )
-                xy_attn_mask = torch.concat([x_attn_mask_pad, y_attn_mask], dim=0).to(
+        xy_attn_mask = torch.concat([x_attn_mask_pad, y_attn_mask], dim=0).to(
-                    y.device
+            x.device
-                )
+        )
-            else:
+        
-                ###最右边一列（是错的）
+
-                # xy_attn_mask=torch.ones((1, x_len+y_len), dtype=torch.bool,device=xy_pos.device)
+        for idx in tqdm(range(1500)):
-                # xy_attn_mask[:,-1]=False
+            
-                ###最下面一行（是对的）
-                xy_attn_mask = torch.zeros(
-                    (1, x_len + y_len), dtype=torch.bool, device=xy_pos.device
-                )
-            # pdb.set_trace()
-            ###缓存重头戏
-            # print(1111,xy_pos.shape,xy_attn_mask.shape,x_len,y_len)
             xy_dec, _ = self.h((xy_pos, None), mask=xy_attn_mask, cache=cache)
             logits = self.ar_predict_layer(
                 xy_dec[:, -1]
@@ -404,6 +398,10 @@ class Text2SemanticDecoder(nn.Module):
             samples = sample(
                 logits[0], y, top_k=top_k, top_p=top_p, repetition_penalty=1.35, temperature=temperature
             )[0].unsqueeze(0)
+            # 本次生成的 semantic_ids 和之前的 y 构成新的 y
+            # print(samples.shape)#[1,1]#第一个1是bs
+            y = torch.concat([y, samples], dim=1) 
+
             if early_stop_num != -1 and (y.shape[1] - prefix_len) > early_stop_num:
                 print("use early stop num:", early_stop_num)
                 stop = True
@@ -412,13 +410,38 @@ class Text2SemanticDecoder(nn.Module):
                 # print(torch.argmax(logits, dim=-1)[0] == self.EOS, samples[0, 0] == self.EOS)
                 stop = True
             if stop:
-                if prompts.shape[1] == y.shape[1]:
+                # if prompts.shape[1] == y.shape[1]:
+                #     y = torch.concat([y, torch.zeros_like(samples)], dim=1)
+                #     print("bad zero prediction")
+                if y.shape[1]==0:
                     y = torch.concat([y, torch.zeros_like(samples)], dim=1)
                     print("bad zero prediction")
                 print(f"T2S Decoding EOS [{prefix_len} -> {y.shape[1]}]")
                 break
-            # 本次生成的 semantic_ids 和之前的 y 构成新的 y
+            
-            # print(samples.shape)#[1,1]#第一个1是bs
+            ####################### update next step ###################################
-            y = torch.concat([y, samples], dim=1)
             cache["first_infer"] = 0
-        return y, idx
+            if cache["y_emb"] is not None:
+                y_emb = torch.cat(
+                    [cache["y_emb"], self.ar_audio_embedding(y[:, -1:])], dim = 1
+                )
+                cache["y_emb"] = y_emb
+                y_pos = self.ar_audio_position(y_emb)
+                xy_pos = y_pos[:, -1:]
+            else:
+                y_emb = self.ar_audio_embedding(y[:, -1:])
+                cache["y_emb"] = y_emb
+                y_pos = self.ar_audio_position(y_emb)
+                xy_pos = y_pos
+            y_len = y_pos.shape[1]
+
+            ###最右边一列（是错的）
+            # xy_attn_mask=torch.ones((1, x_len+y_len), dtype=torch.bool,device=xy_pos.device)
+            # xy_attn_mask[:,-1]=False
+            ###最下面一行（是对的）
+            xy_attn_mask = torch.zeros(
+                (1, x_len + y_len), dtype=torch.bool, device=xy_pos.device
+            )
+        if ref_free:
+            return y[:, :-1], 0
+        return y[:, :-1], idx-1

GPT_SoVITS/AR/models/utils.py

@@ -114,7 +114,8 @@ def logits_to_probs(
     top_p: Optional[int] = None,
     repetition_penalty: float = 1.0,
 ):
-    previous_tokens = previous_tokens.squeeze()
+    if previous_tokens is not None:
+        previous_tokens = previous_tokens.squeeze()
     # print(logits.shape,previous_tokens.shape)
     # pdb.set_trace()
     if previous_tokens is not None and repetition_penalty != 1.0:

GPT_SoVITS/AR/modules/patched_mha_with_cache.py

@@ -5,8 +5,8 @@ from torch.nn.functional import (
     _none_or_dtype,
     _in_projection_packed,
 )
-
+from torch.nn import functional as F
-# import torch
+import torch
 # Tensor = torch.Tensor
 # from typing import Callable, List, Optional, Tuple, Union
 
@@ -448,9 +448,11 @@ def multi_head_attention_forward_patched(
         k = k.view(bsz, num_heads, src_len, head_dim)
         v = v.view(bsz, num_heads, src_len, head_dim)
 
+        # with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=True):
         attn_output = scaled_dot_product_attention(
             q, k, v, attn_mask, dropout_p, is_causal
         )
+
         attn_output = (
             attn_output.permute(2, 0, 1, 3).contiguous().view(bsz * tgt_len, embed_dim)
         )

GPT_SoVITS/inference_webui.py

@@ -365,15 +365,19 @@ def merge_short_text_in_array(texts, threshold):
             result[len(result) - 1] += text
     return result
 
-def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切"), top_k=20, top_p=0.6, temperature=0.6):
+def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language, how_to_cut=i18n("不切"), top_k=20, top_p=0.6, temperature=0.6, ref_free = False):
+    if prompt_text is None or len(prompt_text) == 0:
+        ref_free = True
     t0 = ttime()
     prompt_language = dict_language[prompt_language]
     text_language = dict_language[text_language]
-    prompt_text = prompt_text.strip("\n")
+    if not ref_free:
-    if (prompt_text[-1] not in splits): prompt_text += "。" if prompt_language != "en" else "."
+        prompt_text = prompt_text.strip("\n")
+        if (prompt_text[-1] not in splits): prompt_text += "。" if prompt_language != "en" else "."
+        print(i18n("实际输入的参考文本:"), prompt_text)
     text = text.strip("\n")
     if (text[0] not in splits and len(get_first(text)) < 4): text = "。" + text if text_language != "en" else "." + text
-    print(i18n("实际输入的参考文本:"), prompt_text)
+    
     print(i18n("实际输入的目标文本:"), text)
     zero_wav = np.zeros(
         int(hps.data.sampling_rate * 0.3),
@@ -398,11 +402,10 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
             1, 2
         )  # .float()
         codes = vq_model.extract_latent(ssl_content)
+   
         prompt_semantic = codes[0, 0]
     t1 = ttime()
 
-    phones1, word2ph1, norm_text1=get_cleaned_text_final(prompt_text, prompt_language)
-
     if (how_to_cut == i18n("凑四句一切")):
         text = cut1(text)
     elif (how_to_cut == i18n("凑50字一切")):
@@ -419,7 +422,9 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
     texts = text.split("\n")
     texts = merge_short_text_in_array(texts, 5)
     audio_opt = []
-    bert1=get_bert_final(phones1, word2ph1, norm_text1,prompt_language,device).to(dtype)
+    if not ref_free:
+        phones1, word2ph1, norm_text1=get_cleaned_text_final(prompt_text, prompt_language)
+        bert1=get_bert_final(phones1, word2ph1, norm_text1,prompt_language,device).to(dtype)
 
     for text in texts:
         # 解决输入目标文本的空行导致报错的问题
@@ -429,9 +434,13 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         print(i18n("实际输入的目标文本(每句):"), text)
         phones2, word2ph2, norm_text2 = get_cleaned_text_final(text, text_language)
         bert2 = get_bert_final(phones2, word2ph2, norm_text2, text_language, device).to(dtype)
-        bert = torch.cat([bert1, bert2], 1)
+        if not ref_free:
+            bert = torch.cat([bert1, bert2], 1)
+            all_phoneme_ids = torch.LongTensor(phones1+phones2).to(device).unsqueeze(0)
+        else:
+            bert = bert2
+            all_phoneme_ids = torch.LongTensor(phones2).to(device).unsqueeze(0)
 
-        all_phoneme_ids = torch.LongTensor(phones1 + phones2).to(device).unsqueeze(0)
         bert = bert.to(device).unsqueeze(0)
         all_phoneme_len = torch.tensor([all_phoneme_ids.shape[-1]]).to(device)
         prompt = prompt_semantic.unsqueeze(0).to(device)
@@ -441,7 +450,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
             pred_semantic, idx = t2s_model.model.infer_panel(
                 all_phoneme_ids,
                 all_phoneme_len,
-                prompt,
+                None if ref_free else prompt,
                 bert,
                 # prompt_phone_len=ph_offset,
                 top_k=top_k,
@@ -607,7 +616,10 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
         gr.Markdown(value=i18n("*请上传并填写参考信息"))
         with gr.Row():
             inp_ref = gr.Audio(label=i18n("请上传3~10秒内参考音频，超过会报错！"), type="filepath")
-            prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="")
+            with gr.Column():
+                ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式 无参考文本时该选项无效"), value=False, interactive=True, show_label=True)
+                gr.Markdown("使用无参考文本模式时建议使用微调GPT")
+                prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="")
             prompt_language = gr.Dropdown(
                 label=i18n("参考音频的语种"), choices=[i18n("中文"), i18n("英文"), i18n("日文"), i18n("中英混合"), i18n("日英混合"), i18n("多语种混合")], value=i18n("中文")
             )
@@ -624,6 +636,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
                 interactive=True,
             )
             with gr.Row():
+                gr.Markdown("gpt采样参数(无参考文本时不要太低)：")
                 top_k = gr.Slider(minimum=1,maximum=100,step=1,label=i18n("top_k"),value=5,interactive=True)
                 top_p = gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("top_p"),value=1,interactive=True)
                 temperature = gr.Slider(minimum=0,maximum=1,step=0.05,label=i18n("temperature"),value=1,interactive=True)
@@ -632,7 +645,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
 
         inference_button.click(
             get_tts_wav,
-            [inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut,top_k,top_p,temperature],
+            [inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut, top_k, top_p, temperature, ref_text_free],
             [output],
         )
 

GPT_SoVITS/module/models.py

@@ -228,6 +228,7 @@ class TextEncoder(nn.Module):
         )
 
         y = self.ssl_proj(y * y_mask) * y_mask
+     
         y = self.encoder_ssl(y * y_mask, y_mask)
 
         text_mask = torch.unsqueeze(
@@ -958,11 +959,13 @@ class SynthesizerTrn(nn.Module):
 
     @torch.no_grad()
     def decode(self, codes, text, refer, noise_scale=0.5):
-        refer_lengths = torch.LongTensor([refer.size(2)]).to(refer.device)
+        ge = None
-        refer_mask = torch.unsqueeze(
+        if refer is not None:
-            commons.sequence_mask(refer_lengths, refer.size(2)), 1
+            refer_lengths = torch.LongTensor([refer.size(2)]).to(refer.device)
-        ).to(refer.dtype)
+            refer_mask = torch.unsqueeze(
-        ge = self.ref_enc(refer * refer_mask, refer_mask)
+                commons.sequence_mask(refer_lengths, refer.size(2)), 1
+            ).to(refer.dtype)
+            ge = self.ref_enc(refer * refer_mask, refer_mask)
 
         y_lengths = torch.LongTensor([codes.size(2) * 2]).to(codes.device)
         text_lengths = torch.LongTensor([text.size(-1)]).to(text.device)