Upload app.py

app.py

@@ -1,5 +1,13 @@
-import os,re
-import gradio as gr
+import os,re,logging
+logging.getLogger("markdown_it").setLevel(logging.ERROR)
+logging.getLogger("urllib3").setLevel(logging.ERROR)
+logging.getLogger("httpcore").setLevel(logging.ERROR)
+logging.getLogger("httpx").setLevel(logging.ERROR)
+logging.getLogger("asyncio").setLevel(logging.ERROR)
+
+logging.getLogger("charset_normalizer").setLevel(logging.ERROR)
+logging.getLogger("torchaudio._extension").setLevel(logging.ERROR)
+import pdb
 
 gpt_path = os.environ.get(
     "gpt_path", "models/Taffy/Taffy-e5.ckpt"
@@ -49,7 +57,6 @@ else:
     bert_model = bert_model.to(device)
 
 
-# bert_model=bert_model.to(device)
 def get_bert_feature(text, word2ph):
     with torch.no_grad():
         inputs = tokenizer(text, return_tensors="pt")
@@ -63,15 +70,8 @@ def get_bert_feature(text, word2ph):
         repeat_feature = res[i].repeat(word2ph[i], 1)
         phone_level_feature.append(repeat_feature)
     phone_level_feature = torch.cat(phone_level_feature, dim=0)
-    # if(is_half==True):phone_level_feature=phone_level_feature.half()
     return phone_level_feature.T
 
-
-n_semantic = 1024
-
-dict_s2=torch.load(sovits_path,map_location="cpu")
-hps=dict_s2["config"]
-
 class DictToAttrRecursive(dict):
     def __init__(self, input_dict):
         super().__init__(input_dict)
@@ -99,12 +99,6 @@ class DictToAttrRecursive(dict):
         except KeyError:
             raise AttributeError(f"Attribute {item} not found")
 
-
-hps = DictToAttrRecursive(hps)
-
-hps.model.semantic_frame_rate = "25hz"
-dict_s1 = torch.load(gpt_path, map_location="cpu")
-config = dict_s1["config"]
 ssl_model = cnhubert.get_model()
 if is_half == True:
     ssl_model = ssl_model.half().to(device)
@@ -123,7 +117,8 @@ def change_sovits_weights(sovits_path):
         n_speakers=hps.data.n_speakers,
         **hps.model
     )
-    del vq_model.enc_q
+    if("pretrained"not in sovits_path):
+        del vq_model.enc_q
     if is_half == True:
         vq_model = vq_model.half().to(device)
     else:
@@ -165,10 +160,88 @@ def get_spepc(hps, filename):
     return spec
 
 
-dict_language = {"中文": "zh", "英文": "en", "日文": "ja"}
-
-
-def get_tts_wav(selected_text, prompt_text, prompt_language, text, text_language):
+dict_language={
+    ("中文"):"zh",
+    ("英文"):"en",
+    ("日文"):"ja"
+}
+
+
+def splite_en_inf(sentence, language):
+    pattern = re.compile(r'[a-zA-Z. ]+')
+    textlist = []
+    langlist = []
+    pos = 0
+    for match in pattern.finditer(sentence):
+        start, end = match.span()
+        if start > pos:
+            textlist.append(sentence[pos:start])
+            langlist.append(language)
+        textlist.append(sentence[start:end])
+        langlist.append("en")
+        pos = end
+    if pos < len(sentence):
+        textlist.append(sentence[pos:])
+        langlist.append(language)
+
+    return textlist, langlist
+
+
+def clean_text_inf(text, language):
+    phones, word2ph, norm_text = clean_text(text, language)
+    phones = cleaned_text_to_sequence(phones)
+
+    return phones, word2ph, norm_text
+def get_bert_inf(phones, word2ph, norm_text, language):
+    if language == "zh":
+        bert = get_bert_feature(norm_text, word2ph).to(device)
+    else:
+        bert = torch.zeros(
+            (1024, len(phones)),
+            dtype=torch.float16 if is_half == True else torch.float32,
+        ).to(device)
+
+    return bert
+
+
+def nonen_clean_text_inf(text, language):
+    textlist, langlist = splite_en_inf(text, language)
+    phones_list = []
+    word2ph_list = []
+    norm_text_list = []
+    for i in range(len(textlist)):
+        lang = langlist[i]
+        phones, word2ph, norm_text = clean_text_inf(textlist[i], lang)
+        phones_list.append(phones)
+        if lang == "en" or "ja":
+            pass
+        else:
+            word2ph_list.append(word2ph)
+        norm_text_list.append(norm_text)
+    print(word2ph_list)
+    phones = sum(phones_list, [])
+    word2ph = sum(word2ph_list, [])
+    norm_text = ' '.join(norm_text_list)
+
+    return phones, word2ph, norm_text
+
+
+def nonen_get_bert_inf(text, language):
+    textlist, langlist = splite_en_inf(text, language)
+    print(textlist)
+    print(langlist)
+    bert_list = []
+    for i in range(len(textlist)):
+        text = textlist[i]
+        lang = langlist[i]
+        phones, word2ph, norm_text = clean_text_inf(text, lang)
+        bert = get_bert_inf(phones, word2ph, norm_text, lang)
+        bert_list.append(bert)
+    bert = torch.cat(bert_list, dim=1)
+
+    return bert
+
+def get_tts_wav(selected_text, prompt_text, prompt_language, text, text_language,how_to_cut=("不切")):
     ref_wav_path = text_to_audio_mappings.get(selected_text, "")
     if not ref_wav_path:
         print("Audio file not found for the selected text.")
@@ -201,28 +274,37 @@ def get_tts_wav(selected_text, prompt_text, prompt_language, text, text_language
     t1 = ttime()
     prompt_language = dict_language[prompt_language]
     text_language = dict_language[text_language]
-    phones1, word2ph1, norm_text1 = clean_text(prompt_text, prompt_language)
-    phones1 = cleaned_text_to_sequence(phones1)
-    texts = text.split("\n")
+    
+    if prompt_language == "en":
+        phones1, word2ph1, norm_text1 = clean_text_inf(prompt_text, prompt_language)
+    else:
+        phones1, word2ph1, norm_text1 = nonen_clean_text_inf(prompt_text, prompt_language)
+    if(how_to_cut==("凑五句一切")):text=cut1(text)
+    elif(how_to_cut==("凑50字一切")):text=cut2(text)
+    elif(how_to_cut==("按中文句号。切")):text=cut3(text)
+    elif(how_to_cut==("按英文句号.切")):text=cut4(text)
+    text = text.replace("\n\n","\n").replace("\n\n","\n").replace("\n\n","\n")
+    if(text[-1]not in splits):text+="。"if text_language!="en"else "."
+    texts=text.split("\n")
     audio_opt = []
+    if prompt_language == "en":
+        bert1 = get_bert_inf(phones1, word2ph1, norm_text1, prompt_language)
+    else:
+        bert1 = nonen_get_bert_inf(prompt_text, prompt_language)
 
     for text in texts:
         # 解决输入目标文本的空行导致报错的问题
         if (len(text.strip()) == 0):
             continue
-        phones2, word2ph2, norm_text2 = clean_text(text, text_language)
-        phones2 = cleaned_text_to_sequence(phones2)
-        if prompt_language == "zh":
-            bert1 = get_bert_feature(norm_text1, word2ph1).to(device)
+        if text_language == "en":
+            phones2, word2ph2, norm_text2 = clean_text_inf(text, text_language)
         else:
-            bert1 = torch.zeros(
-                (1024, len(phones1)),
-                dtype=torch.float16 if is_half == True else torch.float32,
-            ).to(device)
-        if text_language == "zh":
-            bert2 = get_bert_feature(norm_text2, word2ph2).to(device)
+            phones2, word2ph2, norm_text2 = nonen_clean_text_inf(text, text_language)
+
+        if text_language == "en":
+            bert2 = get_bert_inf(phones2, word2ph2, norm_text2, text_language)
         else:
-            bert2 = torch.zeros((1024, len(phones2))).to(bert1)
+            bert2 = nonen_get_bert_inf(text, text_language)
         bert = torch.cat([bert1, bert2], 1)
 
         all_phoneme_ids = torch.LongTensor(phones1 + phones2).to(device).unsqueeze(0)
@@ -380,7 +462,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
     ### <center>⚠️在线端不稳定且生成速度较慢，强烈建议下载模型本地推理！\n
                 """)
     # with gr.Tabs():
-    #     with gr.TabItem(i18n("伴奏人声分离&去混响&去回声")):
+
     with gr.Group():
         gr.Markdown(value="*参考音频选择（必选）")
         with gr.Row():