提高切分生成的鲁棒性

GPT_SoVITS/inference_webui.py

@@ -17,6 +17,7 @@ logging.getLogger("charset_normalizer").setLevel(logging.ERROR)
 logging.getLogger("torchaudio._extension").setLevel(logging.ERROR)
 import pdb
 import torch
+import jieba
 
 if os.path.exists("./gweight.txt"):
     with open("./gweight.txt", 'r', encoding="utf-8") as file:
@@ -228,7 +229,7 @@ def get_bert_inf(phones, word2ph, norm_text, language):
 
 
 splits = {"，", "。", "？", "！", ",", ".", "?", "!", "~", ":", "：", "—", "…", }
-
+MAX_LENGTH = 500
 
 def get_first(text):
     pattern = "[" + "".join(re.escape(sep) for sep in splits) + "]"
@@ -432,6 +433,21 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         np.int16
     )
 
+def split_by_tokenizer(text):
+    tokens = jieba.lcut(text)
+
+    slices = []
+    temp_slice = []
+    for token in tokens:
+        if len("".join(temp_slice + [token])) > MAX_LENGTH:
+            slices.append("".join(temp_slice))
+            temp_slice = []
+        temp_slice.append(token)
+
+    if temp_slice:
+        slices.append("".join(temp_slice))
+
+    return slices
 
 def split(todo_text):
     todo_text = todo_text.replace("……", "。").replace("——", "，")
@@ -440,55 +456,63 @@ def split(todo_text):
     i_split_head = i_split_tail = 0
     len_text = len(todo_text)
     todo_texts = []
-    while 1:
+    while True:
         if i_split_head >= len_text:
             break  # 结尾一定有标点，所以直接跳出即可，最后一段在上次已加入
         if todo_text[i_split_head] in splits:
             i_split_head += 1
-            todo_texts.append(todo_text[i_split_tail:i_split_head])
+            segment = todo_text[i_split_tail:i_split_head]
+            if len(segment) > MAX_LENGTH:
+                # 如果段落长度超过最大长度，进行语义切分
+                todo_texts.extend(split_by_tokenizer(segment))
+            else:
+                todo_texts.append(segment)
             i_split_tail = i_split_head
         else:
             i_split_head += 1
     return todo_texts
 
-
+def cut1(inp, max_concat=4):
-def cut1(inp):
     inp = inp.strip("\n")
     inps = split(inp)
-    split_idx = list(range(0, len(inps), 4))
-    split_idx[-1] = None
-    if len(split_idx) > 1:
     opts = []
-        for idx in range(len(split_idx) - 1):
+    temp_segment = []
-            opts.append("".join(inps[split_idx[idx]: split_idx[idx + 1]]))
+
+    for segment in inps:
+        if len("".join(temp_segment + [segment])) > MAX_LENGTH or len(temp_segment) + 1 > max_concat:
+            opts.append("".join(temp_segment))
+            temp_segment = [segment]
         else:
-        opts = [inp]
+            temp_segment.append(segment)
+
+    if temp_segment:
+        opts.append("".join(temp_segment))
+
     return "\n".join(opts)
 
 
-def cut2(inp):
+def cut2(inp, desired_length=50):
     inp = inp.strip("\n")
     inps = split(inp)
     if len(inps) < 2:
         return inp
     opts = []
-    summ = 0
     tmp_str = ""
-    for i in range(len(inps)):
+
-        summ += len(inps[i])
+    for segment in inps:
-        tmp_str += inps[i]
+        if len(tmp_str + segment) > desired_length:
-        if summ > 50:
+            if tmp_str:
-            summ = 0
                 opts.append(tmp_str)
-            tmp_str = ""
+            tmp_str = segment
-    if tmp_str != "":
+        else:
+            tmp_str += segment
+    if tmp_str:
         opts.append(tmp_str)
-    # print(opts)
+    if len(opts) > 1 and len(opts[-1]) < desired_length and len(opts[-1]) + len(opts[-2]) < MAX_LENGTH:  ##如果最后一个太短了，和前一个合一起
-    if len(opts) > 1 and len(opts[-1]) < 50:  ##如果最后一个太短了，和前一个合一起
         opts[-2] = opts[-2] + opts[-1]
         opts = opts[:-1]
-    return "\n".join(opts)
 
+    return "\n".join(opts)
 
 def cut3(inp):
     inp = inp.strip("\n")