Merge remote-tracking branch 'upstream/main' into chores

GPT_SoVITS/inference_webui.py

@@ -311,14 +311,14 @@ def get_phones_and_bert(text,language,version):
         if language == "zh":
             if re.search(r'[A-Za-z]', formattext):
                 formattext = re.sub(r'[a-z]', lambda x: x.group(0).upper(), formattext)
-                formattext = chinese.text_normalize(formattext)
+                formattext = chinese.mix_text_normalize(formattext)
                 return get_phones_and_bert(formattext,"zh",version)
             else:
                 phones, word2ph, norm_text = clean_text_inf(formattext, language, version)
                 bert = get_bert_feature(norm_text, word2ph).to(device)
         elif language == "yue" and re.search(r'[A-Za-z]', formattext):
                 formattext = re.sub(r'[a-z]', lambda x: x.group(0).upper(), formattext)
-                formattext = chinese.text_normalize(formattext)
+                formattext = chinese.mix_text_normalize(formattext)
                 return get_phones_and_bert(formattext,"yue",version)
         else:
             phones, word2ph, norm_text = clean_text_inf(formattext, language, version)

GPT_SoVITS/s2_train.py

@@ -1,5 +1,6 @@
+import warnings
+warnings.filterwarnings("ignore")
 import utils, os
-
 hps = utils.get_hparams(stage=2)
 os.environ["CUDA_VISIBLE_DEVICES"] = hps.train.gpu_numbers.replace("-", ",")
 import torch
@@ -119,7 +120,7 @@ def run(rank, n_gpus, hps):
         collate_fn=collate_fn,
         batch_sampler=train_sampler,
         persistent_workers=True,
-        prefetch_factor=16,
+        prefetch_factor=4,
     )
     # if rank == 0:
     #     eval_dataset = TextAudioSpeakerLoader(hps.data.validation_files, hps.data, val=True)
@@ -211,7 +212,7 @@ def run(rank, n_gpus, hps):
         # traceback.print_exc()
         epoch_str = 1
         global_step = 0
-        if hps.train.pretrained_s2G != "":
+        if hps.train.pretrained_s2G != ""and hps.train.pretrained_s2G != None and os.path.exists(hps.train.pretrained_s2G):
             if rank == 0:
                 logger.info("loaded pretrained %s" % hps.train.pretrained_s2G)
             print(
@@ -223,7 +224,7 @@ def run(rank, n_gpus, hps):
                     strict=False,
                 )
             )  ##测试不加载优化器
-        if hps.train.pretrained_s2D != "":
+        if hps.train.pretrained_s2D != ""and hps.train.pretrained_s2D != None and os.path.exists(hps.train.pretrained_s2D):
             if rank == 0:
                 logger.info("loaded pretrained %s" % hps.train.pretrained_s2D)
             print(

GPT_SoVITS/text/cantonese.py

@@ -1,6 +1,6 @@
+# reference: https://huggingface.co/spaces/Naozumi0512/Bert-VITS2-Cantonese-Yue/blob/main/text/chinese.py
+
 import sys
-# sys.path.append("/data/docker/liujing04/gpt-vits/mq-vits-s1bert_no_bert-kv_cache-new_text-emo")
-sys.path.append("/data/docker/liujing04/gpt-vits/mq-vits-s1bert_no_bert-kv_cache-new_text-emo2v2")
 import re
 import cn2an
 

GPT_SoVITS/text/chinese.py

@@ -47,6 +47,19 @@ def replace_punctuation(text):
 
     replaced_text = pattern.sub(lambda x: rep_map[x.group()], text)
 
+    replaced_text = re.sub(
+        r"[^\u4e00-\u9fa5" + "".join(punctuation) + r"]+", "", replaced_text
+    )
+
+    return replaced_text
+
+
+def replace_punctuation_with_en(text):
+    text = text.replace("嗯", "恩").replace("呣", "母")
+    pattern = re.compile("|".join(re.escape(p) for p in rep_map.keys()))
+
+    replaced_text = pattern.sub(lambda x: rep_map[x.group()], text)
+
     replaced_text = re.sub(
         r"[^\u4e00-\u9fa5A-Za-z" + "".join(punctuation) + r"]+", "", replaced_text
     )
@@ -171,6 +184,20 @@ def text_normalize(text):
     return dest_text
 
 
+# 不排除英文的文本格式化
+def mix_text_normalize(text):
+    # https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/paddlespeech/t2s/frontend/zh_normalization
+    tx = TextNormalizer()
+    sentences = tx.normalize(text)
+    dest_text = ""
+    for sentence in sentences:
+        dest_text += replace_punctuation_with_en(sentence)
+
+    # 避免重复标点引起的参考泄露
+    dest_text = replace_consecutive_punctuation(dest_text)
+    return dest_text
+
+
 if __name__ == "__main__":
     text = "啊——但是《原神》是由,米哈\游自主，研发的一款全.新开放世界.冒险游戏"
     text = "呣呣呣～就是…大人的鼹鼠党吧？"

GPT_SoVITS/text/chinese2.py

@@ -4,6 +4,7 @@ import re
 
 import cn2an
 from pypinyin import lazy_pinyin, Style
+from pypinyin.contrib.tone_convert import to_normal, to_finals_tone3, to_initials, to_finals
 
 from text.symbols import punctuation
 from text.tone_sandhi import ToneSandhi
@@ -24,7 +25,7 @@ import jieba_fast.posseg as psg
 is_g2pw = True#True if is_g2pw_str.lower() == 'true' else False
 if is_g2pw:
     print("当前使用g2pw进行拼音推理")
-    from text.g2pw import G2PWPinyin
+    from text.g2pw import G2PWPinyin, correct_pronunciation
     parent_directory = os.path.dirname(current_file_path)
     g2pw = G2PWPinyin(model_dir="GPT_SoVITS/text/G2PWModel",model_source="GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large",v_to_u=False, neutral_tone_with_five=True)
 
@@ -42,6 +43,8 @@ rep_map = {
     "$": ".",
     "/": ",",
     "—": "-",
+    "~": "…",
+    "～":"…",
 }
 
 tone_modifier = ToneSandhi()
@@ -71,16 +74,10 @@ def _get_initials_finals(word):
     initials = []
     finals = []
 
-    if not is_g2pw:
+    orig_initials = lazy_pinyin(word, neutral_tone_with_five=True, style=Style.INITIALS)
-        orig_initials = lazy_pinyin(word, neutral_tone_with_five=True, style=Style.INITIALS)
+    orig_finals = lazy_pinyin(
-        orig_finals = lazy_pinyin(
+        word, neutral_tone_with_five=True, style=Style.FINALS_TONE3
-            word, neutral_tone_with_five=True, style=Style.FINALS_TONE3
+    )
-        )
-    else:
-        orig_initials = g2pw.lazy_pinyin(word, neutral_tone_with_five=True, style=Style.INITIALS)
-        orig_finals = g2pw.lazy_pinyin(
-            word, neutral_tone_with_five=True, style=Style.FINALS_TONE3
-        )
 
     for c, v in zip(orig_initials, orig_finals):
         initials.append(c)
@@ -88,6 +85,54 @@ def _get_initials_finals(word):
     return initials, finals
 
 
+must_erhua = {
+    "小院儿", "胡同儿", "范儿", "老汉儿", "撒欢儿", "寻老礼儿", "妥妥儿", "媳妇儿"
+}
+not_erhua = {
+    "虐儿", "为儿", "护儿", "瞒儿", "救儿", "替儿", "有儿", "一儿", "我儿", "俺儿", "妻儿",
+    "拐儿", "聋儿", "乞儿", "患儿", "幼儿", "孤儿", "婴儿", "婴幼儿", "连体儿", "脑瘫儿",
+    "流浪儿", "体弱儿", "混血儿", "蜜雪儿", "舫儿", "祖儿", "美儿", "应采儿", "可儿", "侄儿",
+    "孙儿", "侄孙儿", "女儿", "男儿", "红孩儿", "花儿", "虫儿", "马儿", "鸟儿", "猪儿", "猫儿",
+    "狗儿", "少儿"
+}
+def _merge_erhua(initials: list[str],
+                finals: list[str],
+                word: str,
+                pos: str) -> list[list[str]]:
+    """
+    Do erhub.
+    """
+    # fix er1
+    for i, phn in enumerate(finals):
+        if i == len(finals) - 1 and word[i] == "儿" and phn == 'er1':
+            finals[i] = 'er2'
+
+    # 发音
+    if word not in must_erhua and (word in not_erhua or
+                                        pos in {"a", "j", "nr"}):
+        return initials, finals
+
+    # "……" 等情况直接返回
+    if len(finals) != len(word):
+        return initials, finals
+
+    assert len(finals) == len(word)
+
+    # 与前一个字发同音
+    new_initials = []
+    new_finals = []
+    for i, phn in enumerate(finals):
+        if i == len(finals) - 1 and word[i] == "儿" and phn in {
+                "er2", "er5"
+        } and word[-2:] not in not_erhua and new_finals:
+            phn = "er" + new_finals[-1][-1]
+
+        new_initials.append(initials[i])
+        new_finals.append(phn)
+
+    return new_initials, new_finals
+
+
 def _g2p(segments):
     phones_list = []
     word2ph = []
@@ -96,20 +141,62 @@ def _g2p(segments):
         # Replace all English words in the sentence
         seg = re.sub("[a-zA-Z]+", "", seg)
         seg_cut = psg.lcut(seg)
+        seg_cut = tone_modifier.pre_merge_for_modify(seg_cut)
         initials = []
         finals = []
-        seg_cut = tone_modifier.pre_merge_for_modify(seg_cut)
+
-        for word, pos in seg_cut:
+        if not is_g2pw:
-            if pos == "eng":
+            for word, pos in seg_cut:
-                continue
+                if pos == "eng":
-            sub_initials, sub_finals = _get_initials_finals(word)
+                    continue
-            sub_finals = tone_modifier.modified_tone(word, pos, sub_finals)
+                sub_initials, sub_finals = _get_initials_finals(word)
-            initials.append(sub_initials)
+                sub_finals = tone_modifier.modified_tone(word, pos, sub_finals)
-            finals.append(sub_finals)
+                # 儿化
-            # assert len(sub_initials) == len(sub_finals) == len(word)
+                sub_initials, sub_finals = _merge_erhua(sub_initials, sub_finals, word, pos)
-        initials = sum(initials, [])
+                initials.append(sub_initials)
-        finals = sum(finals, [])
+                finals.append(sub_finals)
-        #
+                # assert len(sub_initials) == len(sub_finals) == len(word)
+            initials = sum(initials, [])
+            finals = sum(finals, [])
+            print("pypinyin结果",initials,finals)
+        else:
+            # g2pw采用整句推理
+            pinyins = g2pw.lazy_pinyin(seg, neutral_tone_with_five=True, style=Style.TONE3)
+
+            pre_word_length = 0
+            for word, pos in seg_cut:
+                sub_initials = []
+                sub_finals = []
+                now_word_length = pre_word_length + len(word)
+
+                if pos == 'eng':
+                    pre_word_length = now_word_length
+                    continue
+
+                word_pinyins = pinyins[pre_word_length:now_word_length]
+
+                # 多音字消歧
+                word_pinyins = correct_pronunciation(word,word_pinyins)
+
+                for pinyin in word_pinyins:
+                    if pinyin[0].isalpha():
+                        sub_initials.append(to_initials(pinyin))
+                        sub_finals.append(to_finals_tone3(pinyin,neutral_tone_with_five=True))
+                    else:
+                        sub_initials.append(pinyin)
+                        sub_finals.append(pinyin)
+
+                pre_word_length = now_word_length
+                sub_finals = tone_modifier.modified_tone(word, pos, sub_finals)
+                # 儿化
+                sub_initials, sub_finals = _merge_erhua(sub_initials, sub_finals, word, pos)
+                initials.append(sub_initials)
+                finals.append(sub_finals)
+
+            initials = sum(initials, [])
+            finals = sum(finals, [])
+            # print("g2pw结果",initials,finals)
+
         for c, v in zip(initials, finals):
             raw_pinyin = c + v
             # NOTE: post process for pypinyin outputs
@@ -164,6 +251,24 @@ def _g2p(segments):
     return phones_list, word2ph
 
 
+def replace_punctuation_with_en(text):
+    text = text.replace("嗯", "恩").replace("呣", "母")
+    pattern = re.compile("|".join(re.escape(p) for p in rep_map.keys()))
+
+    replaced_text = pattern.sub(lambda x: rep_map[x.group()], text)
+
+    replaced_text = re.sub(
+        r"[^\u4e00-\u9fa5A-Za-z" + "".join(punctuation) + r"]+", "", replaced_text
+    )
+
+    return replaced_text
+
+def replace_consecutive_punctuation(text):
+    punctuations = ''.join(re.escape(p) for p in punctuation)
+    pattern = f'([{punctuations}])([{punctuations}])+'
+    result = re.sub(pattern, r'\1', text)
+    return result
+
 def text_normalize(text):
     # https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/paddlespeech/t2s/frontend/zh_normalization
     tx = TextNormalizer()
@@ -171,6 +276,22 @@ def text_normalize(text):
     dest_text = ""
     for sentence in sentences:
         dest_text += replace_punctuation(sentence)
+
+    # 避免重复标点引起的参考泄露
+    dest_text = replace_consecutive_punctuation(dest_text)
+    return dest_text
+
+# 不排除英文的文本格式化
+def mix_text_normalize(text):
+    # https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/paddlespeech/t2s/frontend/zh_normalization
+    tx = TextNormalizer()
+    sentences = tx.normalize(text)
+    dest_text = ""
+    for sentence in sentences:
+        dest_text += replace_punctuation_with_en(sentence)
+
+    # 避免重复标点引起的参考泄露
+    dest_text = replace_consecutive_punctuation(dest_text)
     return dest_text
 
 

GPT_SoVITS/text/g2pw/g2pw.py

@@ -13,6 +13,7 @@ from .onnx_api import G2PWOnnxConverter
 current_file_path = os.path.dirname(__file__)
 CACHE_PATH = os.path.join(current_file_path, "polyphonic.pickle")
 PP_DICT_PATH = os.path.join(current_file_path, "polyphonic.rep")
+PP_FIX_DICT_PATH = os.path.join(current_file_path, "polyphonic-fix.rep")
 
 
 class G2PWPinyin(Pinyin):
@@ -69,12 +70,6 @@ class Converter(UltimateConverter):
     def _to_pinyin(self, han, style, heteronym, errors, strict, **kwargs):
         pinyins = []
 
-        if han in pp_dict:
-            phns = pp_dict[han]
-            for ph in phns:
-                pinyins.append([ph])
-            return pinyins
-
         g2pw_pinyin = self._g2pw(han)
 
         if not g2pw_pinyin:  # g2pw 不支持的汉字改为使用 pypinyin 原有逻辑
@@ -139,7 +134,21 @@ def read_dict():
             value = eval(value_str.strip())
             polyphonic_dict[key.strip()] = value
             line = f.readline()
+    with open(PP_FIX_DICT_PATH) as f:
+        line = f.readline()
+        while line:
+            key, value_str = line.split(':')
+            value = eval(value_str.strip())
+            polyphonic_dict[key.strip()] = value
+            line = f.readline()
     return polyphonic_dict
 
 
+def correct_pronunciation(word,word_pinyins):
+    if word in pp_dict:
+        word_pinyins = pp_dict[word]
+
+    return word_pinyins
+
+
 pp_dict = get_dict()

GPT_SoVITS/text/g2pw/onnx_api.py

@@ -1,6 +1,8 @@
 # This code is modified from https://github.com/PaddlePaddle/PaddleSpeech/tree/develop/paddlespeech/t2s/frontend/g2pw
 # This code is modified from https://github.com/GitYCC/g2pW
 
+import warnings
+warnings.filterwarnings("ignore")
 import json
 import os
 import zipfile,requests

GPT_SoVITS/text/korean.py

@@ -1,3 +1,5 @@
+# reference: https://github.com/ORI-Muchim/MB-iSTFT-VITS-Korean/blob/main/text/korean.py
+
 import re
 from jamo import h2j, j2hcj
 import ko_pron

GPT_SoVITS/text/zh_normalization/num.py

@@ -28,7 +28,7 @@ UNITS = OrderedDict({
     8: '亿',
 })
 
-COM_QUANTIFIERS = '(封|艘|把|目|套|段|人|所|朵|匹|张|座|回|场|尾|条|个|首|阙|阵|网|炮|顶|丘|棵|只|支|袭|辆|挑|担|颗|壳|窠|曲|墙|群|腔|砣|座|客|贯|扎|捆|刀|令|打|手|罗|坡|山|岭|江|溪|钟|队|单|双|对|出|口|头|脚|板|跳|枝|件|贴|针|线|管|名|位|身|堂|课|本|页|家|户|层|丝|毫|厘|分|钱|两|斤|担|铢|石|钧|锱|忽|(千|毫|微)克|毫|厘|(公)分|分|寸|尺|丈|里|寻|常|铺|程|(千|分|厘|毫|微)米|米|撮|勺|合|升|斗|石|盘|碗|碟|叠|桶|笼|盆|盒|杯|钟|斛|锅|簋|篮|盘|桶|罐|瓶|壶|卮|盏|箩|箱|煲|啖|袋|钵|年|月|日|季|刻|时|周|天|秒|分|小时|旬|纪|岁|世|更|夜|春|夏|秋|冬|代|伏|辈|丸|泡|粒|颗|幢|堆|条|根|支|道|面|片|张|颗|块|元|(亿|千万|百万|万|千|百)|(亿|千万|百万|万|千|百|美|)元|(亿|千万|百万|万|千|百|十|)吨|(亿|千万|百万|万|千|百|)块|角|毛|分)'
+COM_QUANTIFIERS = '(处|台|架|枚|趟|幅|平|方|堵|间|床|株|批|项|例|列|篇|栋|注|亩|封|艘|把|目|套|段|人|所|朵|匹|张|座|回|场|尾|条|个|首|阙|阵|网|炮|顶|丘|棵|只|支|袭|辆|挑|担|颗|壳|窠|曲|墙|群|腔|砣|座|客|贯|扎|捆|刀|令|打|手|罗|坡|山|岭|江|溪|钟|队|单|双|对|出|口|头|脚|板|跳|枝|件|贴|针|线|管|名|位|身|堂|课|本|页|家|户|层|丝|毫|厘|分|钱|两|斤|担|铢|石|钧|锱|忽|(千|毫|微)克|毫|厘|(公)分|分|寸|尺|丈|里|寻|常|铺|程|(千|分|厘|毫|微)米|米|撮|勺|合|升|斗|石|盘|碗|碟|叠|桶|笼|盆|盒|杯|钟|斛|锅|簋|篮|盘|桶|罐|瓶|壶|卮|盏|箩|箱|煲|啖|袋|钵|年|月|日|季|刻|时|周|天|秒|分|小时|旬|纪|岁|世|更|夜|春|夏|秋|冬|代|伏|辈|丸|泡|粒|颗|幢|堆|条|根|支|道|面|片|张|颗|块|元|(亿|千万|百万|万|千|百)|(亿|千万|百万|万|千|百|美|)元|(亿|千万|百万|万|千|百|十|)吨|(亿|千万|百万|万|千|百|)块|角|毛|分)'
 
 # 分数表达式
 RE_FRAC = re.compile(r'(-?)(\d+)/(\d+)')

docs/cn/Changelog_CN.md

@@ -213,16 +213,24 @@
 
 重点：
 
-4-[支持合成语速调节。支持冻结随机性只调节语速。](https://github.com/RVC-Boss/GPT-SoVITS/commit/9588a3c52d9ebdb20b3c5d74f647d12e7c1171c2)
+4-[支持合成语速调节。支持冻结随机性只调节语速，](https://github.com/RVC-Boss/GPT-SoVITS/commit/9588a3c52d9ebdb20b3c5d74f647d12e7c1171c2)并将其更新到api.py上https://github.com/RVC-Boss/GPT-SoVITS/pull/1340
 
 
-todolist：
+### 20240806
 
-1-中文多音字推理优化(有没有人来测试的，欢迎把测试结果写在pr评论区里) https://github.com/RVC-Boss/GPT-SoVITS/pull/488
+1-增加bs-roformer人声伴奏分离模型支持。 https://github.com/RVC-Boss/GPT-SoVITS/pull/1306 https://github.com/RVC-Boss/GPT-SoVITS/pull/1356 [支持fp16推理。](https://github.com/RVC-Boss/GPT-SoVITS/commit/e62e965323a60a76a025bcaa45268c1ddcbcf05c)
-（v2底模训练已经合了，下个版本发布就要合了）
-
-2-正在尝试解决低音质参考音频导致音质较差的问题(202407:已经解决了)，节点暂定8月吧
 
+2-更好的中文文本前端。 https://github.com/RVC-Boss/GPT-SoVITS/pull/987 https://github.com/RVC-Boss/GPT-SoVITS/pull/1351 https://github.com/RVC-Boss/GPT-SoVITS/pull/1404 优化多音字逻辑（v2版本特供）。 https://github.com/RVC-Boss/GPT-SoVITS/pull/488
+
+3-自动填充下一步的文件路径 https://github.com/RVC-Boss/GPT-SoVITS/pull/1356
+
+4-增加喂饭逻辑，用户瞎写显卡序号也可以正常运作 [#bce451a](https://github.com/RVC-Boss/GPT-SoVITS/commit/bce451a2d1641e581e200297d01f219aeaaf7299) [4c8b761](https://github.com/RVC-Boss/GPT-SoVITS/commit/4c8b7612206536b8b4435997acb69b25d93acb78)
+
+5-增加粤语ASR支持
+
+6-GPT-SoVITS-v2支持
+
+7-计时逻辑优化 https://github.com/RVC-Boss/GPT-SoVITS/pull/1387
 
 
 

go-webui-v1.bat

@@ -0,0 +1,2 @@
+runtime\python.exe webui.py v1 zh_CN
+pause

go-webui-v1.ps1

@@ -1,4 +1,4 @@
 $ErrorActionPreference = "SilentlyContinue"
 chcp 65001
-& "$PSScriptRoot\runtime\python.exe" "$PSScriptRoot\webui.py v2 zh_CN"
+& "$PSScriptRoot\runtime\python.exe" "$PSScriptRoot\webui.py v1 zh_CN"
 pause

go-webui-v2.bat

@@ -1,2 +0,0 @@
-runtime\python.exe webui.py v2 zh_CN
-pause

webui.py

@@ -1,15 +1,15 @@
-import os,shutil,sys,pdb,re
+import os,sys
 if len(sys.argv)==1:sys.argv.append('v2')
 version="v1"if sys.argv[1]=="v1" else"v2"
 os.environ["version"]=version
 now_dir = os.getcwd()
 sys.path.insert(0, now_dir)
-import json,yaml,warnings,torch
+import warnings
+warnings.filterwarnings("ignore")
+import json,yaml,torch,pdb,re,shutil
 import platform
 import psutil
 import signal
-
-warnings.filterwarnings("ignore")
 torch.manual_seed(233333)
 tmp = os.path.join(now_dir, "TEMP")
 os.makedirs(tmp, exist_ok=True)
@@ -112,17 +112,6 @@ def fix_gpu_numbers(inputs):
     except:
         return inputs
 
-def get_weights_names(GPT_weight_root, SoVITS_weight_root):
-    SoVITS_names = [i for i in pretrained_sovits_name]
-    for path in SoVITS_weight_root:
-        for name in os.listdir(path):
-            if name.endswith(".pth"): SoVITS_names.append("%s/%s" % (path, name))
-    GPT_names = [i for i in pretrained_gpt_name]
-    for path in GPT_weight_root:
-        for name in os.listdir(path):
-            if name.endswith(".ckpt"): GPT_names.append("%s/%s" % (path, name))
-    return SoVITS_names, GPT_names
-
 pretrained_sovits_name=["GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s2G2333k.pth", "GPT_SoVITS/pretrained_models/s2G488k.pth"]
 pretrained_gpt_name=["GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s1bert25hz-5kh-longer-epoch=12-step=369668.ckpt", "GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt"]
 
@@ -132,19 +121,33 @@ _=''
 for i in pretrained_model_list:
     if os.path.exists(i):...
     else:_+=f'\n    {i}'
-if _:raise FileExistsError(i18n('以下模型不存在:')+_)
+if _:
+    print("warning:",i18n('以下模型不存在:')+_)
 
 _ =[[],[]]
 for i in range(2):
-    if os.path.exists(pretrained_gpt_name[i]):
+    if os.path.exists(pretrained_gpt_name[i]):_[0].append(pretrained_gpt_name[i])
-        _[0].append(pretrained_gpt_name[i])
+    else:_[0].append("")##没有下pretrained模型的，说不定他们是想自己从零训底模呢
-    if os.path.exists(pretrained_sovits_name[i]):
+    if os.path.exists(pretrained_sovits_name[i]):_[-1].append(pretrained_sovits_name[i])
-        _[-1].append(pretrained_sovits_name[i])
+    else:_[-1].append("")
 pretrained_gpt_name,pretrained_sovits_name = _
 
 SoVITS_weight_root=["SoVITS_weights_v2","SoVITS_weights"]
 GPT_weight_root=["GPT_weights_v2","GPT_weights"]
-SoVITS_names,GPT_names = get_weights_names(GPT_weight_root, SoVITS_weight_root)
+for root in SoVITS_weight_root+GPT_weight_root:
+    os.makedirs(root,exist_ok=True)
+def get_weights_names():
+    SoVITS_names = [name for name in pretrained_sovits_name if name!=""]
+    for path in SoVITS_weight_root:
+        for name in os.listdir(path):
+            if name.endswith(".pth"): SoVITS_names.append("%s/%s" % (path, name))
+    GPT_names = [name for name in pretrained_gpt_name if name!=""]
+    for path in GPT_weight_root:
+        for name in os.listdir(path):
+            if name.endswith(".ckpt"): GPT_names.append("%s/%s" % (path, name))
+    return SoVITS_names, GPT_names
+
+SoVITS_names,GPT_names = get_weights_names()
 for path in SoVITS_weight_root+GPT_weight_root:
     os.makedirs(path,exist_ok=True)
 
@@ -315,8 +318,9 @@ def open1Ba(batch_size,total_epoch,exp_name,text_low_lr_rate,if_save_latest,if_s
         data["train"]["if_save_every_weights"]=if_save_every_weights
         data["train"]["save_every_epoch"]=save_every_epoch
         data["train"]["gpu_numbers"]=gpu_numbers1Ba
+        data["model"]["version"]=version
         data["data"]["exp_dir"]=data["s2_ckpt_dir"]=s2_dir
-        data["save_weight_dir"]=SoVITS_weight_root
+        data["save_weight_dir"]=SoVITS_weight_root[-int(version[-1])+2]
         data["name"]=exp_name
         data["version"]=version
         tmp_config_path="%s/tmp_s2.json"%tmp
@@ -358,12 +362,12 @@ def open1Bb(batch_size,total_epoch,exp_name,if_dpo,if_save_latest,if_save_every_
         data["train"]["if_save_every_weights"]=if_save_every_weights
         data["train"]["if_save_latest"]=if_save_latest
         data["train"]["if_dpo"]=if_dpo
-        data["train"]["half_weights_save_dir"]=GPT_weight_root
+        data["train"]["half_weights_save_dir"]=GPT_weight_root[-int(version[-1])+2]
         data["train"]["exp_name"]=exp_name
         data["train_semantic_path"]="%s/6-name2semantic.tsv"%s1_dir
         data["train_phoneme_path"]="%s/2-name2text.txt"%s1_dir
         data["output_dir"]="%s/logs_s1"%s1_dir
-        data["version"]=version
+        # data["version"]=version
 
         os.environ["_CUDA_VISIBLE_DEVICES"]=fix_gpu_numbers(gpu_numbers.replace("-",","))
         os.environ["hz"]="25hz"
@@ -866,10 +870,10 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             with gr.Row():
                 exp_name = gr.Textbox(label=i18n("*实验/模型名"), value="xxx", interactive=True)
                 gpu_info = gr.Textbox(label=i18n("显卡信息"), value=gpu_info, visible=True, interactive=False)
-                version_checkbox = gr.Radio(label=i18n("版本"),value='v2',choices=['v1','v2'])
+                version_checkbox = gr.Radio(label=i18n("版本"),value=version,choices=['v1','v2'])
-                pretrained_s2G = gr.Textbox(label=i18n("预训练的SoVITS-G模型路径"), value=pretrained_sovits_name[0], interactive=True)
+                pretrained_s2G = gr.Textbox(label=i18n("预训练的SoVITS-G模型路径"), value=pretrained_sovits_name[-int(version[-1])+2], interactive=True)
-                pretrained_s2D = gr.Textbox(label=i18n("预训练的SoVITS-D模型路径"), value=pretrained_sovits_name[0].replace("s2G","s2D"), interactive=True)
+                pretrained_s2D = gr.Textbox(label=i18n("预训练的SoVITS-D模型路径"), value=pretrained_sovits_name[-int(version[-1])+2].replace("s2G","s2D"), interactive=True)
-                pretrained_s1 = gr.Textbox(label=i18n("预训练的GPT模型路径"), value=pretrained_gpt_name[0], interactive=True)
+                pretrained_s1 = gr.Textbox(label=i18n("预训练的GPT模型路径"), value=pretrained_gpt_name[-int(version[-1])+2], interactive=True)
             with gr.TabItem(i18n("1A-训练集格式化工具")):
                 gr.Markdown(value=i18n("输出logs/实验名目录下应有23456开头的文件和文件夹"))
                 with gr.Row():