Merge 2e230d055efc7a37e908a3b28bec8914792397e7 into fdf794e31d1fd6f91c5cb4fbb0396094491a31ac

Update WSL Rocm (#2561 )
fix: 更正链接 (#2539 )
2026-07-11 18:03:18 +08:00 · 2025-08-09 02:21:03 +00:00 · 2025-08-02 17:47:15 +08:00 · 2025-07-19 00:29:48 +08:00 · 2025-07-18 14:59:59 +08:00 · 2025-07-18 11:54:40 +08:00
26 changed files with 846 additions and 377 deletions
--- a/GPT_SoVITS/AR/models/t2s_model.py
+++ b/GPT_SoVITS/AR/models/t2s_model.py
@ -356,7 +356,7 @@ class Text2SemanticDecoder(nn.Module):
        x = self.ar_text_embedding(x)
        x = x + self.bert_proj(bert_feature.transpose(1, 2))
        x = self.ar_text_position(x)
-        x_mask = make_pad_mask(x_lens)
+        x_mask = make_pad_mask_left(x_lens)

        y_mask = make_pad_mask(y_lens)
        y_mask_int = y_mask.type(torch.int64)
@ -420,7 +420,7 @@ class Text2SemanticDecoder(nn.Module):
            mask=xy_attn_mask,
        )
        x_len = x_lens.max()
-        logits = self.ar_predict_layer(xy_dec[:, x_len:])
+        logits = self.ar_predict_layer(xy_dec[:, x_len-1:])

        ###### DPO #############
        reject_xy_pos, reject_xy_attn_mask, reject_targets = self.make_input_data(
@ -432,7 +432,7 @@ class Text2SemanticDecoder(nn.Module):
            mask=reject_xy_attn_mask,
        )
        x_len = x_lens.max()
-        reject_logits = self.ar_predict_layer(reject_xy_dec[:, x_len:])
+        reject_logits = self.ar_predict_layer(reject_xy_dec[:, x_len-1:])

        # loss
        # from feiteng: 每次 duration 越多, 梯度更新也应该更多, 所以用 sum
@ -455,7 +455,7 @@ class Text2SemanticDecoder(nn.Module):
        x = self.ar_text_embedding(x)
        x = x + self.bert_proj(bert_feature.transpose(1, 2))
        x = self.ar_text_position(x)
-        x_mask = make_pad_mask(x_lens)
+        x_mask = make_pad_mask_left(x_lens)

        y_mask = make_pad_mask(y_lens)
        y_mask_int = y_mask.type(torch.int64)
@ -502,7 +502,7 @@ class Text2SemanticDecoder(nn.Module):
            (xy_pos, None),
            mask=xy_attn_mask,
        )
-        logits = self.ar_predict_layer(xy_dec[:, x_len:]).permute(0, 2, 1)
+        logits = self.ar_predict_layer(xy_dec[:, x_len-1:]).permute(0, 2, 1)
        # loss
        # from feiteng: 每次 duration 越多, 梯度更新也应该更多, 所以用 sum
        loss = F.cross_entropy(logits, targets, reduction="sum")
@ -578,7 +578,7 @@ class Text2SemanticDecoder(nn.Module):
    def pad_y_eos(self, y, y_mask_int, eos_id):
        targets = F.pad(y, (0, 1), value=0) + eos_id * F.pad(y_mask_int, (0, 1), value=1)
        # 错位
-        return targets[:, :-1], targets[:, 1:]
+        return targets[:, :-1], targets

    def infer_panel_batch_infer(
        self,
--- a/GPT_SoVITS/TTS_infer_pack/TTS.py
+++ b/GPT_SoVITS/TTS_infer_pack/TTS.py
@ -304,10 +304,10 @@ class TTS_Config:
            configs: dict = self._load_configs(self.configs_path)

        assert isinstance(configs, dict)
-        version = configs.get("version", "v2").lower()
-        assert version in ["v1", "v2", "v3", "v4", "v2Pro", "v2ProPlus"]
-        self.default_configs[version] = configs.get(version, self.default_configs[version])
-        self.configs: dict = configs.get("custom", deepcopy(self.default_configs[version]))
+        configs_ = deepcopy(self.default_configs)
+        configs_.update(configs)
+        self.configs: dict = configs_.get("custom", configs_["v2"])
+        self.default_configs = deepcopy(configs_)

        self.device = self.configs.get("device", torch.device("cpu"))
        if "cuda" in str(self.device) and not torch.cuda.is_available():
@ -315,11 +315,13 @@ class TTS_Config:
            self.device = torch.device("cpu")

        self.is_half = self.configs.get("is_half", False)
-        # if str(self.device) == "cpu" and self.is_half:
-        #     print(f"Warning: Half precision is not supported on CPU, set is_half to False.")
-        #     self.is_half = False
+        if str(self.device) == "cpu" and self.is_half:
+            print(f"Warning: Half precision is not supported on CPU, set is_half to False.")
+            self.is_half = False

+        version = self.configs.get("version", None)
        self.version = version
+        assert self.version in ["v1", "v2", "v3", "v4", "v2Pro", "v2ProPlus"], "Invalid version!"
        self.t2s_weights_path = self.configs.get("t2s_weights_path", None)
        self.vits_weights_path = self.configs.get("vits_weights_path", None)
        self.bert_base_path = self.configs.get("bert_base_path", None)
@ -576,6 +578,10 @@ class TTS:
        if self.configs.is_half and str(self.configs.device) != "cpu":
            self.vits_model = self.vits_model.half()

+        self.configs.save_configs()
+
+
+
    def init_t2s_weights(self, weights_path: str):
        print(f"Loading Text2Semantic weights from {weights_path}")
        self.configs.t2s_weights_path = weights_path
--- a/GPT_SoVITS/TTS_infer_pack/TextPreprocessor.py
+++ b/GPT_SoVITS/TTS_infer_pack/TextPreprocessor.py
@ -121,71 +121,67 @@ class TextPreprocessor:

    def get_phones_and_bert(self, text: str, language: str, version: str, final: bool = False):
        with self.bert_lock:
-            if language in {"en", "all_zh", "all_ja", "all_ko", "all_yue"}:
-                # language = language.replace("all_","")
-                formattext = text
-                while "  " in formattext:
-                    formattext = formattext.replace("  ", " ")
-                if language == "all_zh":
-                    if re.search(r"[A-Za-z]", formattext):
-                        formattext = re.sub(r"[a-z]", lambda x: x.group(0).upper(), formattext)
-                        formattext = chinese.mix_text_normalize(formattext)
-                        return self.get_phones_and_bert(formattext, "zh", version)
+            text = re.sub(r' {2,}', ' ', text)
+            textlist = []
+            langlist = []
+            if language == "all_zh":
+                for tmp in LangSegmenter.getTexts(text,"zh"):
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+            elif language == "all_yue":
+                for tmp in LangSegmenter.getTexts(text,"zh"):
+                    if tmp["lang"] == "zh":
+                        tmp["lang"] = "yue"
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+            elif language == "all_ja":
+                for tmp in LangSegmenter.getTexts(text,"ja"):
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+            elif language == "all_ko":
+                for tmp in LangSegmenter.getTexts(text,"ko"):
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+            elif language == "en":
+                langlist.append("en")
+                textlist.append(text)
+            elif language == "auto":
+                for tmp in LangSegmenter.getTexts(text):
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+            elif language == "auto_yue":
+                for tmp in LangSegmenter.getTexts(text):
+                    if tmp["lang"] == "zh":
+                        tmp["lang"] = "yue"
+                    langlist.append(tmp["lang"])
+                    textlist.append(tmp["text"])
+            else:
+                for tmp in LangSegmenter.getTexts(text):
+                    if langlist:
+                        if (tmp["lang"] == "en" and langlist[-1] == "en") or (tmp["lang"] != "en" and langlist[-1] != "en"):
+                            textlist[-1] += tmp["text"]
+                            continue
+                    if tmp["lang"] == "en":
+                        langlist.append(tmp["lang"])
                    else:
-                        phones, word2ph, norm_text = self.clean_text_inf(formattext, language, version)
-                        bert = self.get_bert_feature(norm_text, word2ph).to(self.device)
-                elif language == "all_yue" and re.search(r"[A-Za-z]", formattext):
-                    formattext = re.sub(r"[a-z]", lambda x: x.group(0).upper(), formattext)
-                    formattext = chinese.mix_text_normalize(formattext)
-                    return self.get_phones_and_bert(formattext, "yue", version)
-                else:
-                    phones, word2ph, norm_text = self.clean_text_inf(formattext, language, version)
-                    bert = torch.zeros(
-                        (1024, len(phones)),
-                        dtype=torch.float32,
-                    ).to(self.device)
-            elif language in {"zh", "ja", "ko", "yue", "auto", "auto_yue"}:
-                textlist = []
-                langlist = []
-                if language == "auto":
-                    for tmp in LangSegmenter.getTexts(text):
-                        langlist.append(tmp["lang"])
-                        textlist.append(tmp["text"])
-                elif language == "auto_yue":
-                    for tmp in LangSegmenter.getTexts(text):
-                        if tmp["lang"] == "zh":
-                            tmp["lang"] = "yue"
-                        langlist.append(tmp["lang"])
-                        textlist.append(tmp["text"])
-                else:
-                    for tmp in LangSegmenter.getTexts(text):
-                        if langlist:
-                            if (tmp["lang"] == "en" and langlist[-1] == "en") or (
-                                tmp["lang"] != "en" and langlist[-1] != "en"
-                            ):
-                                textlist[-1] += tmp["text"]
-                                continue
-                        if tmp["lang"] == "en":
-                            langlist.append(tmp["lang"])
-                        else:
-                            # 因无法区别中日韩文汉字,以用户输入为准
-                            langlist.append(language)
-                        textlist.append(tmp["text"])
-                # print(textlist)
-                # print(langlist)
-                phones_list = []
-                bert_list = []
-                norm_text_list = []
-                for i in range(len(textlist)):
-                    lang = langlist[i]
-                    phones, word2ph, norm_text = self.clean_text_inf(textlist[i], lang, version)
-                    bert = self.get_bert_inf(phones, word2ph, norm_text, lang)
-                    phones_list.append(phones)
-                    norm_text_list.append(norm_text)
-                    bert_list.append(bert)
-                bert = torch.cat(bert_list, dim=1)
-                phones = sum(phones_list, [])
-                norm_text = "".join(norm_text_list)
+                        # 因无法区别中日韩文汉字,以用户输入为准
+                        langlist.append(language)
+                    textlist.append(tmp["text"])
+            # print(textlist)
+            # print(langlist)
+            phones_list = []
+            bert_list = []
+            norm_text_list = []
+            for i in range(len(textlist)):
+                lang = langlist[i]
+                phones, word2ph, norm_text = self.clean_text_inf(textlist[i], lang, version)
+                bert = self.get_bert_inf(phones, word2ph, norm_text, lang)
+                phones_list.append(phones)
+                norm_text_list.append(norm_text)
+                bert_list.append(bert)
+            bert = torch.cat(bert_list, dim=1)
+            phones = sum(phones_list, [])
+            norm_text = "".join(norm_text_list)

            if not final and len(phones) < 6:
                return self.get_phones_and_bert("." + text, language, version, final=True)
@ -240,4 +236,4 @@ class TextPreprocessor:
        punctuations = "".join(re.escape(p) for p in punctuation)
        pattern = f"([{punctuations}])([{punctuations}])+"
        result = re.sub(pattern, r"\1", text)
-        return result
+        return result
--- a/GPT_SoVITS/configs/tts_infer.yaml
+++ b/GPT_SoVITS/configs/tts_infer.yaml
@ -22,6 +22,22 @@ v2:
  t2s_weights_path: GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s1bert25hz-5kh-longer-epoch=12-step=369668.ckpt
  version: v2
  vits_weights_path: GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s2G2333k.pth
+v2Pro:
+  bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
+  cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
+  device: cpu
+  is_half: false
+  t2s_weights_path: GPT_SoVITS/pretrained_models/s1v3.ckpt
+  version: v2Pro
+  vits_weights_path: GPT_SoVITS/pretrained_models/v2Pro/s2Gv2Pro.pth
+v2ProPlus:
+  bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
+  cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
+  device: cpu
+  is_half: false
+  t2s_weights_path: GPT_SoVITS/pretrained_models/s1v3.ckpt
+  version: v2ProPlus
+  vits_weights_path: GPT_SoVITS/pretrained_models/v2Pro/s2Gv2ProPlus.pth
 v3:
  bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
  cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
--- a/GPT_SoVITS/export_torch_script.py
+++ b/GPT_SoVITS/export_torch_script.py
@ -474,6 +474,10 @@ class T2SModel(nn.Module):
        bert = bert.unsqueeze(0)

        x = self.ar_text_embedding(all_phoneme_ids)
+
+        # avoid dtype inconsistency when exporting
+        bert = bert.to(dtype=self.bert_proj.weight.dtype)
+        
        x = x + self.bert_proj(bert.transpose(1, 2))
        x: torch.Tensor = self.ar_text_position(x)

--- a/GPT_SoVITS/inference_webui.py
+++ b/GPT_SoVITS/inference_webui.py
@ -6,7 +6,20 @@
 全部按英文识别
 全部按日文识别
 """
+import psutil
+import os

+def set_high_priority():
+    """把当前 Python 进程设为 HIGH_PRIORITY_CLASS"""
+    if os.name != "nt":
+        return # 仅 Windows 有效
+    p = psutil.Process(os.getpid())
+    try:
+        p.nice(psutil.HIGH_PRIORITY_CLASS)
+        print("已将进程优先级设为 High")
+    except psutil.AccessDenied:
+        print("权限不足，无法修改优先级（请用管理员运行）")
+set_high_priority()
 import json
 import logging
 import os
@ -586,68 +599,67 @@ from text import chinese


 def get_phones_and_bert(text, language, version, final=False):
-    if language in {"en", "all_zh", "all_ja", "all_ko", "all_yue"}:
-        formattext = text
-        while "  " in formattext:
-            formattext = formattext.replace("  ", " ")
-        if language == "all_zh":
-            if re.search(r"[A-Za-z]", formattext):
-                formattext = re.sub(r"[a-z]", lambda x: x.group(0).upper(), formattext)
-                formattext = chinese.mix_text_normalize(formattext)
-                return get_phones_and_bert(formattext, "zh", version)
+    text = re.sub(r' {2,}', ' ', text)
+    textlist = []
+    langlist = []
+    if language == "all_zh":
+        for tmp in LangSegmenter.getTexts(text,"zh"):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "all_yue":
+        for tmp in LangSegmenter.getTexts(text,"zh"):
+            if tmp["lang"] == "zh":
+                tmp["lang"] = "yue"
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "all_ja":
+        for tmp in LangSegmenter.getTexts(text,"ja"):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "all_ko":
+        for tmp in LangSegmenter.getTexts(text,"ko"):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "en":
+        langlist.append("en")
+        textlist.append(text)
+    elif language == "auto":
+        for tmp in LangSegmenter.getTexts(text):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "auto_yue":
+        for tmp in LangSegmenter.getTexts(text):
+            if tmp["lang"] == "zh":
+                tmp["lang"] = "yue"
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    else:
+        for tmp in LangSegmenter.getTexts(text):
+            if langlist:
+                if (tmp["lang"] == "en" and langlist[-1] == "en") or (tmp["lang"] != "en" and langlist[-1] != "en"):
+                    textlist[-1] += tmp["text"]
+                    continue
+            if tmp["lang"] == "en":
+                langlist.append(tmp["lang"])
            else:
-                phones, word2ph, norm_text = clean_text_inf(formattext, language, version)
-                bert = get_bert_feature(norm_text, word2ph).to(device)
-        elif language == "all_yue" and re.search(r"[A-Za-z]", formattext):
-            formattext = re.sub(r"[a-z]", lambda x: x.group(0).upper(), 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)
-            bert = torch.zeros(
-                (1024, len(phones)),
-                dtype=torch.float16 if is_half == True else torch.float32,
-            ).to(device)
-    elif language in {"zh", "ja", "ko", "yue", "auto", "auto_yue"}:
-        textlist = []
-        langlist = []
-        if language == "auto":
-            for tmp in LangSegmenter.getTexts(text):
-                langlist.append(tmp["lang"])
-                textlist.append(tmp["text"])
-        elif language == "auto_yue":
-            for tmp in LangSegmenter.getTexts(text):
-                if tmp["lang"] == "zh":
-                    tmp["lang"] = "yue"
-                langlist.append(tmp["lang"])
-                textlist.append(tmp["text"])
-        else:
-            for tmp in LangSegmenter.getTexts(text):
-                if langlist:
-                    if (tmp["lang"] == "en" and langlist[-1] == "en") or (tmp["lang"] != "en" and langlist[-1] != "en"):
-                        textlist[-1] += tmp["text"]
-                        continue
-                if tmp["lang"] == "en":
-                    langlist.append(tmp["lang"])
-                else:
-                    # 因无法区别中日韩文汉字,以用户输入为准
-                    langlist.append(language)
-                textlist.append(tmp["text"])
-        print(textlist)
-        print(langlist)
-        phones_list = []
-        bert_list = []
-        norm_text_list = []
-        for i in range(len(textlist)):
-            lang = langlist[i]
-            phones, word2ph, norm_text = clean_text_inf(textlist[i], lang, version)
-            bert = get_bert_inf(phones, word2ph, norm_text, lang)
-            phones_list.append(phones)
-            norm_text_list.append(norm_text)
-            bert_list.append(bert)
-        bert = torch.cat(bert_list, dim=1)
-        phones = sum(phones_list, [])
-        norm_text = "".join(norm_text_list)
+                # 因无法区别中日韩文汉字,以用户输入为准
+                langlist.append(language)
+            textlist.append(tmp["text"])
+    print(textlist)
+    print(langlist)
+    phones_list = []
+    bert_list = []
+    norm_text_list = []
+    for i in range(len(textlist)):
+        lang = langlist[i]
+        phones, word2ph, norm_text = clean_text_inf(textlist[i], lang, version)
+        bert = get_bert_inf(phones, word2ph, norm_text, lang)
+        phones_list.append(phones)
+        norm_text_list.append(norm_text)
+        bert_list.append(bert)
+    bert = torch.cat(bert_list, dim=1)
+    phones = sum(phones_list, [])
+    norm_text = "".join(norm_text_list)

    if not final and len(phones) < 6:
        return get_phones_and_bert("." + text, language, version, final=True)
--- a/GPT_SoVITS/inference_webui_fast.py
+++ b/GPT_SoVITS/inference_webui_fast.py
@ -6,7 +6,20 @@
 全部按英文识别
 全部按日文识别
 """
+import psutil
+import os

+def set_high_priority():
+    """把当前 Python 进程设为 HIGH_PRIORITY_CLASS"""
+    if os.name != "nt":
+        return # 仅 Windows 有效
+    p = psutil.Process(os.getpid())
+    try:
+        p.nice(psutil.HIGH_PRIORITY_CLASS)
+        print("已将进程优先级设为 High")
+    except psutil.AccessDenied:
+        print("权限不足，无法修改优先级（请用管理员运行）")
+set_high_priority()
 import json
 import logging
 import os
@ -112,7 +125,8 @@ is_exist_s2gv4 = os.path.exists(path_sovits_v4)
 tts_config = TTS_Config("GPT_SoVITS/configs/tts_infer.yaml")
 tts_config.device = device
 tts_config.is_half = is_half
-tts_config.version = version
+# tts_config.version = version
+tts_config.update_version(version)
 if gpt_path is not None:
    if "！" in gpt_path or "!" in gpt_path:
        gpt_path = name2gpt_path[gpt_path]
--- a/GPT_SoVITS/text/LangSegmenter/langsegmenter.py
+++ b/GPT_SoVITS/text/LangSegmenter/langsegmenter.py
@ -3,44 +3,38 @@ import re

 # jieba静音
 import jieba
-
 jieba.setLogLevel(logging.CRITICAL)

 # 更改fast_langdetect大模型位置
 from pathlib import Path
 import fast_langdetect
-
-fast_langdetect.infer._default_detector = fast_langdetect.infer.LangDetector(
-    fast_langdetect.infer.LangDetectConfig(
-        cache_dir=Path(__file__).parent.parent.parent / "pretrained_models" / "fast_langdetect"
-    )
-)
+fast_langdetect.infer._default_detector = fast_langdetect.infer.LangDetector(fast_langdetect.infer.LangDetectConfig(cache_dir=Path(__file__).parent.parent.parent / "pretrained_models" / "fast_langdetect"))


 from split_lang import LangSplitter


 def full_en(text):
-    pattern = r"^(?=.*[A-Za-z])[A-Za-z0-9\s\u0020-\u007E\u2000-\u206F\u3000-\u303F\uFF00-\uFFEF]+$"
+    pattern = r'^(?=.*[A-Za-z])[A-Za-z0-9\s\u0020-\u007E\u2000-\u206F\u3000-\u303F\uFF00-\uFFEF]+$'
    return bool(re.match(pattern, text))


 def full_cjk(text):
    # 来自wiki
    cjk_ranges = [
-        (0x4E00, 0x9FFF),  # CJK Unified Ideographs
-        (0x3400, 0x4DB5),  # CJK Extension A
-        (0x20000, 0x2A6DD),  # CJK Extension B
-        (0x2A700, 0x2B73F),  # CJK Extension C
-        (0x2B740, 0x2B81F),  # CJK Extension D
-        (0x2B820, 0x2CEAF),  # CJK Extension E
-        (0x2CEB0, 0x2EBEF),  # CJK Extension F
-        (0x30000, 0x3134A),  # CJK Extension G
-        (0x31350, 0x323AF),  # CJK Extension H
-        (0x2EBF0, 0x2EE5D),  # CJK Extension H
+        (0x4E00, 0x9FFF),        # CJK Unified Ideographs
+        (0x3400, 0x4DB5),        # CJK Extension A
+        (0x20000, 0x2A6DD),      # CJK Extension B
+        (0x2A700, 0x2B73F),      # CJK Extension C
+        (0x2B740, 0x2B81F),      # CJK Extension D
+        (0x2B820, 0x2CEAF),      # CJK Extension E
+        (0x2CEB0, 0x2EBEF),      # CJK Extension F
+        (0x30000, 0x3134A),      # CJK Extension G
+        (0x31350, 0x323AF),      # CJK Extension H
+        (0x2EBF0, 0x2EE5D),      # CJK Extension H
    ]

-    pattern = r"[0-9、-〜。！？.!?… /]+$"
+    pattern = r'[0-9、-〜。！？.!?… /]+$'

    cjk_text = ""
    for char in text:
@ -51,7 +45,7 @@ def full_cjk(text):
    return cjk_text


-def split_jako(tag_lang, item):
+def split_jako(tag_lang,item):
    if tag_lang == "ja":
        pattern = r"([\u3041-\u3096\u3099\u309A\u30A1-\u30FA\u30FC]+(?:[0-9、-〜。！？.!?… ]+[\u3041-\u3096\u3099\u309A\u30A1-\u30FA\u30FC]*)*)"
    else:
@ -59,118 +53,165 @@ def split_jako(tag_lang, item):

    lang_list: list[dict] = []
    tag = 0
-    for match in re.finditer(pattern, item["text"]):
+    for match in re.finditer(pattern, item['text']):
        if match.start() > tag:
-            lang_list.append({"lang": item["lang"], "text": item["text"][tag : match.start()]})
+            lang_list.append({'lang':item['lang'],'text':item['text'][tag:match.start()]})

        tag = match.end()
-        lang_list.append({"lang": tag_lang, "text": item["text"][match.start() : match.end()]})
+        lang_list.append({'lang':tag_lang,'text':item['text'][match.start():match.end()]})

-    if tag < len(item["text"]):
-        lang_list.append({"lang": item["lang"], "text": item["text"][tag : len(item["text"])]})
+    if tag < len(item['text']):
+        lang_list.append({'lang':item['lang'],'text':item['text'][tag:len(item['text'])]})

    return lang_list


 def merge_lang(lang_list, item):
-    if lang_list and item["lang"] == lang_list[-1]["lang"]:
-        lang_list[-1]["text"] += item["text"]
+    if lang_list and item['lang'] == lang_list[-1]['lang']:
+        lang_list[-1]['text'] += item['text']
    else:
        lang_list.append(item)
    return lang_list


-class LangSegmenter:
+class LangSegmenter():
    # 默认过滤器, 基于gsv目前四种语言
    DEFAULT_LANG_MAP = {
        "zh": "zh",
        "yue": "zh",  # 粤语
        "wuu": "zh",  # 吴语
        "zh-cn": "zh",
-        "zh-tw": "x",  # 繁体设置为x
+        "zh-tw": "x", # 繁体设置为x
        "ko": "ko",
        "ja": "ja",
        "en": "en",
    }

-    def getTexts(text):
+    def getTexts(text,default_lang = ""):
        lang_splitter = LangSplitter(lang_map=LangSegmenter.DEFAULT_LANG_MAP)
+        lang_splitter.merge_across_digit = False
        substr = lang_splitter.split_by_lang(text=text)

        lang_list: list[dict] = []

-        for _, item in enumerate(substr):
-            dict_item = {"lang": item.lang, "text": item.text}
+        have_num = False

-            # 处理短英文被识别为其他语言的问题
-            if full_en(dict_item["text"]):
-                dict_item["lang"] = "en"
-                lang_list = merge_lang(lang_list, dict_item)
+        for _, item in enumerate(substr):
+            dict_item = {'lang':item.lang,'text':item.text}
+
+            if dict_item['lang'] == 'digit':
+                if default_lang != "":
+                    dict_item['lang'] = default_lang
+                else:
+                    have_num = True
+                lang_list = merge_lang(lang_list,dict_item)
                continue

-            # 处理非日语夹日文的问题(不包含CJK)
-            ja_list: list[dict] = []
-            if dict_item["lang"] != "ja":
-                ja_list = split_jako("ja", dict_item)
+            # 处理短英文被识别为其他语言的问题
+            if full_en(dict_item['text']):  
+                dict_item['lang'] = 'en'
+                lang_list = merge_lang(lang_list,dict_item)
+                continue

-            if not ja_list:
-                ja_list.append(dict_item)
+            if default_lang != "":
+                dict_item['lang'] = default_lang
+                lang_list = merge_lang(lang_list,dict_item)
+                continue
+            else:
+                # 处理非日语夹日文的问题(不包含CJK)
+                ja_list: list[dict] = []
+                if dict_item['lang'] != 'ja':
+                    ja_list = split_jako('ja',dict_item)

-            # 处理非韩语夹韩语的问题(不包含CJK)
-            ko_list: list[dict] = []
-            temp_list: list[dict] = []
-            for _, ko_item in enumerate(ja_list):
-                if ko_item["lang"] != "ko":
-                    ko_list = split_jako("ko", ko_item)
+                if not ja_list:
+                    ja_list.append(dict_item)

-                if ko_list:
-                    temp_list.extend(ko_list)
-                else:
-                    temp_list.append(ko_item)
+                # 处理非韩语夹韩语的问题(不包含CJK)
+                ko_list: list[dict] = []
+                temp_list: list[dict] = []
+                for _, ko_item in enumerate(ja_list):
+                    if ko_item["lang"] != 'ko':
+                        ko_list = split_jako('ko',ko_item)

-            # 未存在非日韩文夹日韩文
-            if len(temp_list) == 1:
-                # 未知语言检查是否为CJK
-                if dict_item["lang"] == "x":
-                    cjk_text = full_cjk(dict_item["text"])
-                    if cjk_text:
-                        dict_item = {"lang": "zh", "text": cjk_text}
-                        lang_list = merge_lang(lang_list, dict_item)
+                    if ko_list:
+                        temp_list.extend(ko_list)
                    else:
-                        lang_list = merge_lang(lang_list, dict_item)
-                    continue
-                else:
-                    lang_list = merge_lang(lang_list, dict_item)
-                    continue
+                        temp_list.append(ko_item)

-            # 存在非日韩文夹日韩文
-            for _, temp_item in enumerate(temp_list):
-                # 未知语言检查是否为CJK
-                if temp_item["lang"] == "x":
-                    cjk_text = full_cjk(dict_item["text"])
-                    if cjk_text:
-                        dict_item = {"lang": "zh", "text": cjk_text}
-                        lang_list = merge_lang(lang_list, dict_item)
+                # 未存在非日韩文夹日韩文
+                if len(temp_list) == 1:
+                    # 未知语言检查是否为CJK
+                    if dict_item['lang'] == 'x':
+                        cjk_text = full_cjk(dict_item['text'])
+                        if cjk_text:
+                            dict_item = {'lang':'zh','text':cjk_text}
+                            lang_list = merge_lang(lang_list,dict_item)
+                        else:
+                            lang_list = merge_lang(lang_list,dict_item)
+                        continue
                    else:
-                        lang_list = merge_lang(lang_list, dict_item)
-                else:
-                    lang_list = merge_lang(lang_list, temp_item)
+                        lang_list = merge_lang(lang_list,dict_item)
+                        continue

+                # 存在非日韩文夹日韩文
+                for _, temp_item in enumerate(temp_list):
+                    # 未知语言检查是否为CJK
+                    if temp_item['lang'] == 'x':
+                        cjk_text = full_cjk(temp_item['text'])
+                        if cjk_text:
+                            lang_list = merge_lang(lang_list,{'lang':'zh','text':cjk_text})
+                        else:
+                            lang_list = merge_lang(lang_list,temp_item)
+                    else:
+                        lang_list = merge_lang(lang_list,temp_item)
+
+        # 有数字
+        if have_num:
+            temp_list = lang_list
+            lang_list = []
+            for i, temp_item in enumerate(temp_list):
+                if temp_item['lang'] == 'digit':
+                    if default_lang:
+                        temp_item['lang'] = default_lang
+                    elif lang_list and i == len(temp_list) - 1:
+                        temp_item['lang'] = lang_list[-1]['lang']
+                    elif not lang_list and i < len(temp_list) - 1:
+                        temp_item['lang'] = temp_list[1]['lang']
+                    elif lang_list and i < len(temp_list) - 1:
+                        if lang_list[-1]['lang'] == temp_list[i + 1]['lang']:
+                            temp_item['lang'] = lang_list[-1]['lang']
+                        elif lang_list[-1]['text'][-1] in [",",".","!","?","，","。","！","？"]:
+                            temp_item['lang'] = temp_list[i + 1]['lang']
+                        elif temp_list[i + 1]['text'][0] in [",",".","!","?","，","。","！","？"]:
+                            temp_item['lang'] = lang_list[-1]['lang']
+                        elif temp_item['text'][-1] in ["。","."]:
+                            temp_item['lang'] = lang_list[-1]['lang']
+                        elif len(lang_list[-1]['text']) >= len(temp_list[i + 1]['text']):
+                            temp_item['lang'] = lang_list[-1]['lang']
+                        else:
+                            temp_item['lang'] = temp_list[i + 1]['lang']
+                    else:
+                        temp_item['lang'] = 'zh'
+
+                lang_list = merge_lang(lang_list,temp_item)
+
+
+        # 筛X
        temp_list = lang_list
        lang_list = []
        for _, temp_item in enumerate(temp_list):
-            if temp_item["lang"] == "x":
+            if temp_item['lang'] == 'x':
                if lang_list:
-                    temp_item["lang"] = lang_list[-1]["lang"]
+                    temp_item['lang'] = lang_list[-1]['lang']
                elif len(temp_list) > 1:
-                    temp_item["lang"] = temp_list[1]["lang"]
+                    temp_item['lang'] = temp_list[1]['lang']
                else:
-                    temp_item["lang"] = "zh"
+                    temp_item['lang'] = 'zh'

-            lang_list = merge_lang(lang_list, temp_item)
+            lang_list = merge_lang(lang_list,temp_item)

        return lang_list
-
+    

 if __name__ == "__main__":
    text = "MyGO?,你也喜欢まいご吗？"
@ -178,3 +219,7 @@ if __name__ == "__main__":

    text = "ねえ、知ってる？最近、僕は天文学を勉強してるんだ。君の瞳が星空みたいにキラキラしてるからさ。"
    print(LangSegmenter.getTexts(text))
+
+    text = "当时ThinkPad T60刚刚发布，一同推出的还有一款名为Advanced Dock的扩展坞配件。这款扩展坞通过连接T60底部的插槽，扩展出包括PCIe在内的一大堆接口，并且自带电源，让T60可以安装桌面显卡来提升性能。"
+    print(LangSegmenter.getTexts(text,"zh"))
+    print(LangSegmenter.getTexts(text))
--- a/GPT_SoVITS/text/chinese.py
+++ b/GPT_SoVITS/text/chinese.py
@ -181,20 +181,6 @@ 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？"
--- a/GPT_SoVITS/text/chinese2.py
+++ b/GPT_SoVITS/text/chinese2.py
@ -326,20 +326,6 @@ 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？"
--- a/GPT_SoVITS/text/g2pw/onnx_api.py
+++ b/GPT_SoVITS/text/g2pw/onnx_api.py
@ -93,13 +93,13 @@ class G2PWOnnxConverter:
        sess_options.graph_optimization_level = onnxruntime.GraphOptimizationLevel.ORT_ENABLE_ALL
        sess_options.execution_mode = onnxruntime.ExecutionMode.ORT_SEQUENTIAL
        sess_options.intra_op_num_threads = 2 if torch.cuda.is_available() else 0
-        try:
+        if "CUDAExecutionProvider" in onnxruntime.get_available_providers():
            self.session_g2pW = onnxruntime.InferenceSession(
                os.path.join(uncompress_path, "g2pW.onnx"),
                sess_options=sess_options,
                providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
            )
-        except:
+        else:
            self.session_g2pW = onnxruntime.InferenceSession(
                os.path.join(uncompress_path, "g2pW.onnx"),
                sess_options=sess_options,
--- a/GPT_SoVITS/text/zh_normalization/num.py
+++ b/GPT_SoVITS/text/zh_normalization/num.py
@ -256,6 +256,24 @@ def replace_to_range(match) -> str:
    return result


+RE_VERSION_NUM = re.compile(r"((\d+)(\.\d+)(\.\d+)?(\.\d+)+)")
+def replace_vrsion_num(match) -> str:
+    """
+    Args:
+        match (re.Match)
+    Returns:
+        str
+    """
+    result = ""
+    for c in match.group(1):
+        if c == ".":
+            result += "点"
+        else:
+            result += num2str(c)
+    return result
+
+
+
 def _get_value(value_string: str, use_zero: bool = True) -> List[str]:
    stripped = value_string.lstrip("0")
    if len(stripped) == 0:
@ -308,7 +326,11 @@ def num2str(value_string: str) -> str:

    result = verbalize_cardinal(integer)

-    decimal = decimal.rstrip("0")
+    if decimal.endswith("0"):
+        decimal = decimal.rstrip("0") + "0"
+    else:
+        decimal = decimal.rstrip("0")
+
    if decimal:
        # '.22' is verbalized as '零点二二'
        # '3.20' is verbalized as '三点二
--- a/GPT_SoVITS/text/zh_normalization/text_normlization.py
+++ b/GPT_SoVITS/text/zh_normalization/text_normlization.py
@ -25,6 +25,7 @@ from .chronology import replace_time
 from .constants import F2H_ASCII_LETTERS
 from .constants import F2H_DIGITS
 from .constants import F2H_SPACE
+from .num import RE_VERSION_NUM
 from .num import RE_DECIMAL_NUM
 from .num import RE_DEFAULT_NUM
 from .num import RE_FRAC
@ -36,6 +37,7 @@ from .num import RE_RANGE
 from .num import RE_TO_RANGE
 from .num import RE_ASMD
 from .num import RE_POWER
+from .num import replace_vrsion_num
 from .num import replace_default_num
 from .num import replace_frac
 from .num import replace_negative_num
@ -158,6 +160,7 @@ class TextNormalizer:
        sentence = RE_RANGE.sub(replace_range, sentence)

        sentence = RE_INTEGER.sub(replace_negative_num, sentence)
+        sentence = RE_VERSION_NUM.sub(replace_vrsion_num, sentence)
        sentence = RE_DECIMAL_NUM.sub(replace_number, sentence)
        sentence = RE_POSITIVE_QUANTIFIERS.sub(replace_positive_quantifier, sentence)
        sentence = RE_DEFAULT_NUM.sub(replace_default_num, sentence)
--- a/README.md
+++ b/README.md
@ -9,16 +9,19 @@ A Powerful Few-shot Voice Conversion and Text-to-Speech WebUI.<br><br>

 <!-- img src="https://counter.seku.su/cmoe?name=gptsovits&theme=r34" /><br> -->

+[![Python](https://img.shields.io/badge/python-3.10--3.12-blue?style=for-the-badge&logo=python)](https://www.python.org)
+[![GitHub release](https://img.shields.io/github/v/release/RVC-Boss/gpt-sovits?style=for-the-badge&logo=github)](https://github.com/RVC-Boss/gpt-sovits/releases)
+
 [![Train In Colab](https://img.shields.io/badge/Colab-Training-F9AB00?style=for-the-badge&logo=googlecolab)](https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main/Colab-WebUI.ipynb)
-[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)
-[![Huggingface](https://img.shields.io/badge/HuggingFace-demo-blue.svg?style=for-the-badge&logo=huggingface)](https://huggingface.co/spaces/lj1995/GPT-SoVITS-v2)
+[![Huggingface](https://img.shields.io/badge/免费在线体验-free_online_demo-yellow.svg?style=for-the-badge&logo=huggingface)](https://lj1995-gpt-sovits-proplus.hf.space/)
 [![Image Size](https://img.shields.io/docker/image-size/xxxxrt666/gpt-sovits/latest?style=for-the-badge&logo=docker)](https://hub.docker.com/r/xxxxrt666/gpt-sovits)

 [![简体中文](https://img.shields.io/badge/简体中文-阅读文档-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e)
 [![English](https://img.shields.io/badge/English-Read%20Docs-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://rentry.co/GPT-SoVITS-guide#/)
 [![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/en/Changelog_EN.md)
+[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)

-**English** | [**中文简体**](./docs/cn/README.md) | [**日本語**](./docs/ja/README.md) | [**한국어**](./docs/ko/README.md) | [**Türkçe**](./docs/tr/README.md)
+**English** | [**简体中文**](./docs/cn/README.md) | [**日本語**](./docs/ja/README.md) | [**한국어**](./docs/ko/README.md) | [**Türkçe**](./docs/tr/README.md)

 </div>

@ -40,6 +43,11 @@ Unseen speakers few-shot fine-tuning demo:

 https://github.com/RVC-Boss/GPT-SoVITS/assets/129054828/05bee1fa-bdd8-4d85-9350-80c060ab47fb

+**RTF(inference speed) of GPT-SoVITS v2 ProPlus**:
+0.028 tested in 4060Ti, 0.014 tested in 4090 (1400words~=4min, inference time is 3.36s), 0.526 in M4 CPU. You can test our [huggingface demo](https://lj1995-gpt-sovits-proplus.hf.space/) (half H200) to experience high-speed inference .
+
+请不要尬黑GPT-SoVITS推理速度慢，谢谢！
+
 **User guide: [简体中文](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e) | [English](https://rentry.co/GPT-SoVITS-guide#/)**

 ## Installation
@ -64,6 +72,14 @@ If you are a Windows user (tested with win>=10), you can [download the integrate

 **Users in China can [download the package here](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e/dkxgpiy9zb96hob4#KTvnO).**

+Install the program by running the following commands:
+
+```pwsh
+conda create -n GPTSoVits python=3.10
+conda activate GPTSoVits
+pwsh -F install.ps1 --Device <CU126|CU128|CPU> --Source <HF|HF-Mirror|ModelScope> [--DownloadUVR5]
+```
+
 ### Linux

 ```bash
--- a/api.py
+++ b/api.py
@ -543,66 +543,65 @@ from text import chinese


 def get_phones_and_bert(text, language, version, final=False):
-    if language in {"en", "all_zh", "all_ja", "all_ko", "all_yue"}:
-        formattext = text
-        while "  " in formattext:
-            formattext = formattext.replace("  ", " ")
-        if language == "all_zh":
-            if re.search(r"[A-Za-z]", formattext):
-                formattext = re.sub(r"[a-z]", lambda x: x.group(0).upper(), formattext)
-                formattext = chinese.mix_text_normalize(formattext)
-                return get_phones_and_bert(formattext, "zh", version)
+    text = re.sub(r' {2,}', ' ', text)
+    textlist = []
+    langlist = []
+    if language == "all_zh":
+        for tmp in LangSegmenter.getTexts(text,"zh"):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "all_yue":
+        for tmp in LangSegmenter.getTexts(text,"zh"):
+            if tmp["lang"] == "zh":
+                tmp["lang"] = "yue"
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "all_ja":
+        for tmp in LangSegmenter.getTexts(text,"ja"):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "all_ko":
+        for tmp in LangSegmenter.getTexts(text,"ko"):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "en":
+        langlist.append("en")
+        textlist.append(text)
+    elif language == "auto":
+        for tmp in LangSegmenter.getTexts(text):
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    elif language == "auto_yue":
+        for tmp in LangSegmenter.getTexts(text):
+            if tmp["lang"] == "zh":
+                tmp["lang"] = "yue"
+            langlist.append(tmp["lang"])
+            textlist.append(tmp["text"])
+    else:
+        for tmp in LangSegmenter.getTexts(text):
+            if langlist:
+                if (tmp["lang"] == "en" and langlist[-1] == "en") or (tmp["lang"] != "en" and langlist[-1] != "en"):
+                    textlist[-1] += tmp["text"]
+                    continue
+            if tmp["lang"] == "en":
+                langlist.append(tmp["lang"])
            else:
-                phones, word2ph, norm_text = clean_text_inf(formattext, language, version)
-                bert = get_bert_feature(norm_text, word2ph).to(device)
-        elif language == "all_yue" and re.search(r"[A-Za-z]", formattext):
-            formattext = re.sub(r"[a-z]", lambda x: x.group(0).upper(), 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)
-            bert = torch.zeros(
-                (1024, len(phones)),
-                dtype=torch.float16 if is_half == True else torch.float32,
-            ).to(device)
-    elif language in {"zh", "ja", "ko", "yue", "auto", "auto_yue"}:
-        textlist = []
-        langlist = []
-        if language == "auto":
-            for tmp in LangSegmenter.getTexts(text):
-                langlist.append(tmp["lang"])
-                textlist.append(tmp["text"])
-        elif language == "auto_yue":
-            for tmp in LangSegmenter.getTexts(text):
-                if tmp["lang"] == "zh":
-                    tmp["lang"] = "yue"
-                langlist.append(tmp["lang"])
-                textlist.append(tmp["text"])
-        else:
-            for tmp in LangSegmenter.getTexts(text):
-                if langlist:
-                    if (tmp["lang"] == "en" and langlist[-1] == "en") or (tmp["lang"] != "en" and langlist[-1] != "en"):
-                        textlist[-1] += tmp["text"]
-                        continue
-                if tmp["lang"] == "en":
-                    langlist.append(tmp["lang"])
-                else:
-                    # 因无法区别中日韩文汉字,以用户输入为准
-                    langlist.append(language)
-                textlist.append(tmp["text"])
-        phones_list = []
-        bert_list = []
-        norm_text_list = []
-        for i in range(len(textlist)):
-            lang = langlist[i]
-            phones, word2ph, norm_text = clean_text_inf(textlist[i], lang, version)
-            bert = get_bert_inf(phones, word2ph, norm_text, lang)
-            phones_list.append(phones)
-            norm_text_list.append(norm_text)
-            bert_list.append(bert)
-        bert = torch.cat(bert_list, dim=1)
-        phones = sum(phones_list, [])
-        norm_text = "".join(norm_text_list)
+                # 因无法区别中日韩文汉字,以用户输入为准
+                langlist.append(language)
+            textlist.append(tmp["text"])
+    phones_list = []
+    bert_list = []
+    norm_text_list = []
+    for i in range(len(textlist)):
+        lang = langlist[i]
+        phones, word2ph, norm_text = clean_text_inf(textlist[i], lang, version)
+        bert = get_bert_inf(phones, word2ph, norm_text, lang)
+        phones_list.append(phones)
+        norm_text_list.append(norm_text)
+        bert_list.append(bert)
+    bert = torch.cat(bert_list, dim=1)
+    phones = sum(phones_list, [])
+    norm_text = "".join(norm_text_list)

    if not final and len(phones) < 6:
        return get_phones_and_bert("." + text, language, version, final=True)
--- a/docs/cn/Changelog_CN.md
+++ b/docs/cn/Changelog_CN.md
@ -594,3 +594,33 @@
  - 内容: 修复实验名结尾出现空格在win中路径不正确的问题
  - 类型: 修复
  - 提交: RVC-Boss
+- 2025.06.10 [Commit#746cb536](https://github.com/RVC-Boss/GPT-SoVITS/commit/746cb536c68b1fe6ce3ca7e882235375b8a8dd89)
+  - 内容: 语种分割优化
+  - 类型: 优化
+  - 提交: KamioRinn
+- 2025.06.11 [Commit#dd2b9253](https://github.com/RVC-Boss/GPT-SoVITS/commit/dd2b9253aabb09db32db7a3344570ed9df043351)
+  - 内容: 修复并行推理对v2pro支持bug
+  - 类型: 修复
+  - 提交: YYuX-1145
+- 2025.06.11 [Commit#ed89a023](https://github.com/RVC-Boss/GPT-SoVITS/commit/ed89a023378dabba9d4b6580235bb9742245816d)
+  - 内容: v2pro对ge提取时会出现数值溢出的问题修复
+  - 类型: 修复
+  - 提交: RVC-Boss
+- 2025.06.11 [Commit#37f5abfc](https://github.com/RVC-Boss/GPT-SoVITS/commit/6fdc67ca83418306f11e90b9139278313ac5c3e9)[Commit#6fdc67ca](https://github.com/RVC-Boss/GPT-SoVITS/commit/37f5abfcb4a6553652235909db2e124b6f8ff3a5)
+  - 内容: install.sh逻辑优化
+  - 类型: 优化
+  - 提交: XXXXRT666
+- 2025.06.27 [Commit#90ebefa7](https://github.com/RVC-Boss/GPT-SoVITS/commit/90ebefa78fd544da36eebe0b2003620879c921b0)
+  - 内容: onnxruntime加载逻辑优化（对gpu/cpu的判断）
+  - 类型: 优化
+  - 提交: KamioRinn
+- 2025.06.27 [Commit#6df61f58](https://github.com/RVC-Boss/GPT-SoVITS/commit/6df61f58e4d18d4c2ad9d1eddd6a1bd690034c23)
+  - 内容: 语言分割及格式化优化
+  - 类型: 优化
+  - 提交: KamioRinn
+- 2025.07.10 [Commit#426e1a2bb](https://github.com/RVC-Boss/GPT-SoVITS/commit/426e1a2bb43614af2479b877c37acfb0591e952f)
+  - 内容: 提升推理进程优先级（修复win11下可能GPU利用率受限的问题）
+  - 类型: 修复
+  - 提交: XianYue0125
+
+
--- a/docs/cn/README.md
+++ b/docs/cn/README.md
@ -7,16 +7,20 @@

 <a href="https://trendshift.io/repositories/7033" target="_blank"><img src="https://trendshift.io/api/badge/repositories/7033" alt="RVC-Boss%2FGPT-SoVITS | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>

+[![Python](https://img.shields.io/badge/python-3.10--3.12-blue?style=for-the-badge&logo=python)](https://www.python.org)
+[![GitHub release](https://img.shields.io/github/v/release/RVC-Boss/gpt-sovits?style=for-the-badge&logo=github)](https://github.com/RVC-Boss/gpt-sovits/releases)
+
 [![Train In Colab](https://img.shields.io/badge/Colab-Training-F9AB00?style=for-the-badge&logo=googlecolab)](https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main/Colab-WebUI.ipynb)
-[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)
-[![Huggingface](https://img.shields.io/badge/HuggingFace-demo-blue.svg?style=for-the-badge&logo=huggingface)](https://huggingface.co/spaces/lj1995/GPT-SoVITS-v2)
+[![Huggingface](https://img.shields.io/badge/免费在线体验-free_online_demo-yellow.svg?style=for-the-badge&logo=huggingface)](https://lj1995-gpt-sovits-proplus.hf.space/)
 [![Image Size](https://img.shields.io/docker/image-size/xxxxrt666/gpt-sovits/latest?style=for-the-badge&logo=docker)](https://hub.docker.com/r/xxxxrt666/gpt-sovits)

 [![简体中文](https://img.shields.io/badge/简体中文-阅读文档-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e)
 [![English](https://img.shields.io/badge/English-Read%20Docs-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://rentry.co/GPT-SoVITS-guide#/)
-[![Change Log](https://img.shields.io/badge/更新日志-查看更新-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/cn/Changelog_CN.md)
+[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/en/Changelog_EN.md)
+[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)

-[**English**](../../README.md) | **中文简体** | [**日本語**](../ja/README.md) | [**한국어**](../ko/README.md) | [**Türkçe**](../tr/README.md)
+
+[**English**](../../README.md) | **简体中文** | [**日本語**](../ja/README.md) | [**한국어**](../ko/README.md) | [**Türkçe**](../tr/README.md)

 </div>

@ -62,6 +66,12 @@

 **中国地区的用户可以[在此处下载整合包](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e/dkxgpiy9zb96hob4#KTvnO).**

+```pwsh
+conda create -n GPTSoVits python=3.10
+conda activate GPTSoVits
+pwsh -F install.ps1 --Device <CU126|CU128|CPU> --Source <HF|HF-Mirror|ModelScope> [--DownloadUVR5]
+```
+
 ### Linux

 ```bash
--- a/docs/ja/README.md
+++ b/docs/ja/README.md
@ -7,16 +7,20 @@

 <a href="https://trendshift.io/repositories/7033" target="_blank"><img src="https://trendshift.io/api/badge/repositories/7033" alt="RVC-Boss%2FGPT-SoVITS | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>

+[![Python](https://img.shields.io/badge/python-3.10--3.12-blue?style=for-the-badge&logo=python)](https://www.python.org)
+[![GitHub release](https://img.shields.io/github/v/release/RVC-Boss/gpt-sovits?style=for-the-badge&logo=github)](https://github.com/RVC-Boss/gpt-sovits/releases)
+
 [![Train In Colab](https://img.shields.io/badge/Colab-Training-F9AB00?style=for-the-badge&logo=googlecolab)](https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main/Colab-WebUI.ipynb)
-[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)
-[![Huggingface](https://img.shields.io/badge/HuggingFace-demo-blue.svg?style=for-the-badge&logo=huggingface)](https://huggingface.co/spaces/lj1995/GPT-SoVITS-v2)
+[![Huggingface](https://img.shields.io/badge/免费在线体验-free_online_demo-yellow.svg?style=for-the-badge&logo=huggingface)](https://lj1995-gpt-sovits-proplus.hf.space/)
 [![Image Size](https://img.shields.io/docker/image-size/xxxxrt666/gpt-sovits/latest?style=for-the-badge&logo=docker)](https://hub.docker.com/r/xxxxrt666/gpt-sovits)

 [![简体中文](https://img.shields.io/badge/简体中文-阅读文档-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e)
 [![English](https://img.shields.io/badge/English-Read%20Docs-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://rentry.co/GPT-SoVITS-guide#/)
-[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/ja/Changelog_JA.md)
+[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/en/Changelog_EN.md)
+[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)

-[**English**](../../README.md) | [**中文简体**](../cn/README.md) | **日本語** | [**한국어**](../ko/README.md) | [**Türkçe**](../tr/README.md)
+
+[**English**](../../README.md) | [**简体中文**](../cn/README.md) | **日本語** | [**한국어**](../ko/README.md) | [**Türkçe**](../tr/README.md)

 </div>

--- a/docs/ko/README.md
+++ b/docs/ko/README.md
@ -7,16 +7,20 @@

 <a href="https://trendshift.io/repositories/7033" target="_blank"><img src="https://trendshift.io/api/badge/repositories/7033" alt="RVC-Boss%2FGPT-SoVITS | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>

+[![Python](https://img.shields.io/badge/python-3.10--3.12-blue?style=for-the-badge&logo=python)](https://www.python.org)
+[![GitHub release](https://img.shields.io/github/v/release/RVC-Boss/gpt-sovits?style=for-the-badge&logo=github)](https://github.com/RVC-Boss/gpt-sovits/releases)
+
 [![Train In Colab](https://img.shields.io/badge/Colab-Training-F9AB00?style=for-the-badge&logo=googlecolab)](https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main/Colab-WebUI.ipynb)
-[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)
-[![Huggingface](https://img.shields.io/badge/HuggingFace-demo-blue.svg?style=for-the-badge&logo=huggingface)](https://huggingface.co/spaces/lj1995/GPT-SoVITS-v2)
+[![Huggingface](https://img.shields.io/badge/免费在线体验-free_online_demo-yellow.svg?style=for-the-badge&logo=huggingface)](https://lj1995-gpt-sovits-proplus.hf.space/)
 [![Image Size](https://img.shields.io/docker/image-size/xxxxrt666/gpt-sovits/latest?style=for-the-badge&logo=docker)](https://hub.docker.com/r/xxxxrt666/gpt-sovits)

 [![简体中文](https://img.shields.io/badge/简体中文-阅读文档-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e)
 [![English](https://img.shields.io/badge/English-Read%20Docs-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://rentry.co/GPT-SoVITS-guide#/)
-[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/ko/Changelog_KO.md)
+[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/en/Changelog_EN.md)
+[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)

-[**English**](../../README.md) | [**中文简体**](../cn/README.md) | [**日本語**](../ja/README.md) | **한국어** | [**Türkçe**](../tr/README.md)
+
+[**English**](../../README.md) | [**简体中文**](../cn/README.md) | [**日本語**](../ja/README.md) | **한국어** | [**Türkçe**](../tr/README.md)

 </div>

@ -58,6 +62,12 @@ https://github.com/RVC-Boss/GPT-SoVITS/assets/129054828/05bee1fa-bdd8-4d85-9350-

 Windows 사용자라면 (win>=10에서 테스트됨), [통합 패키지를 다운로드](https://huggingface.co/lj1995/GPT-SoVITS-windows-package/resolve/main/GPT-SoVITS-v3lora-20250228.7z?download=true)한 후 압축을 풀고 _go-webui.bat_ 파일을 더블 클릭하면 GPT-SoVITS-WebUI를 시작할 수 있습니다.

+```pwsh
+conda create -n GPTSoVits python=3.10
+conda activate GPTSoVits
+pwsh -F install.ps1 --Device <CU126|CU128|CPU> --Source <HF|HF-Mirror|ModelScope> [--DownloadUVR5]
+```
+
 ### Linux

 ```bash
--- a/docs/tr/README.md
+++ b/docs/tr/README.md
@ -7,16 +7,19 @@ Güçlü Birkaç Örnekli Ses Dönüştürme ve Metinden Konuşmaya Web Arayüz

 <a href="https://trendshift.io/repositories/7033" target="_blank"><img src="https://trendshift.io/api/badge/repositories/7033" alt="RVC-Boss%2FGPT-SoVITS | Trendshift" style="width: 250px; height: 55px;" width="250" height="55"/></a>

+[![Python](https://img.shields.io/badge/python-3.10--3.12-blue?style=for-the-badge&logo=python)](https://www.python.org)
+[![GitHub release](https://img.shields.io/github/v/release/RVC-Boss/gpt-sovits?style=for-the-badge&logo=github)](https://github.com/RVC-Boss/gpt-sovits/releases)
+
 [![Train In Colab](https://img.shields.io/badge/Colab-Training-F9AB00?style=for-the-badge&logo=googlecolab)](https://colab.research.google.com/github/RVC-Boss/GPT-SoVITS/blob/main/Colab-WebUI.ipynb)
-[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)
-[![Huggingface](https://img.shields.io/badge/HuggingFace-demo-blue.svg?style=for-the-badge&logo=huggingface)](https://huggingface.co/spaces/lj1995/GPT-SoVITS-v2)
+[![Huggingface](https://img.shields.io/badge/免费在线体验-free_online_demo-yellow.svg?style=for-the-badge&logo=huggingface)](https://lj1995-gpt-sovits-proplus.hf.space/)
 [![Image Size](https://img.shields.io/docker/image-size/xxxxrt666/gpt-sovits/latest?style=for-the-badge&logo=docker)](https://hub.docker.com/r/xxxxrt666/gpt-sovits)

 [![简体中文](https://img.shields.io/badge/简体中文-阅读文档-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e)
 [![English](https://img.shields.io/badge/English-Read%20Docs-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://rentry.co/GPT-SoVITS-guide#/)
-[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/tr/Changelog_TR.md)
+[![Change Log](https://img.shields.io/badge/Change%20Log-View%20Updates-blue?style=for-the-badge&logo=googledocs&logoColor=white)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/docs/en/Changelog_EN.md)
+[![License](https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative)](https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE)

-[**English**](../../README.md) | [**中文简体**](../cn/README.md) | [**日本語**](../ja/README.md) | [**한국어**](../ko/README.md) | **Türkçe**
+[**English**](../../README.md) | [**简体中文**](../cn/README.md) | [**日本語**](../ja/README.md) | [**한국어**](../ko/README.md) | **Türkçe**

 </div>

@ -58,6 +61,12 @@ https://github.com/RVC-Boss/GPT-SoVITS/assets/129054828/05bee1fa-bdd8-4d85-9350-

 Eğer bir Windows kullanıcısıysanız (win>=10 ile test edilmiştir), [entegre paketi indirin](https://huggingface.co/lj1995/GPT-SoVITS-windows-package/resolve/main/GPT-SoVITS-v3lora-20250228.7z?download=true) ve _go-webui.bat_ dosyasına çift tıklayarak GPT-SoVITS-WebUI'yi başlatın.

+```pwsh
+conda create -n GPTSoVits python=3.10
+conda activate GPTSoVits
+pwsh -F install.ps1 --Device <CU126|CU128|CPU> --Source <HF|HF-Mirror|ModelScope> [--DownloadUVR5]
+```
+
 ### Linux

 ```bash
--- a/install.ps1
+++ b/install.ps1
@ -0,0 +1,241 @@
+Param (
+    [Parameter(Mandatory=$true)][ValidateSet("CU126", "CU128", "CPU")][string]$Device,
+    [Parameter(Mandatory=$true)][ValidateSet("HF", "HF-Mirror", "ModelScope")][string]$Source,
+    [switch]$DownloadUVR5
+)
+
+$global:ErrorActionPreference = 'Stop'
+
+trap {
+    Write-ErrorLog $_
+}
+
+function Write-ErrorLog {
+    param (
+        [System.Management.Automation.ErrorRecord]$ErrorRecord
+    )
+
+    Write-Host "`n[ERROR] Command failed:" -ForegroundColor Red
+    if (-not $ErrorRecord.Exception.Message){
+    } else {
+        Write-Host "Message:" -ForegroundColor Red 
+        $ErrorRecord.Exception.Message -split "`n" | ForEach-Object {
+            Write-Host "    $_"
+        }
+    }
+
+    Write-Host "Command:" -ForegroundColor Red  -NoNewline
+    Write-Host " $($ErrorRecord.InvocationInfo.Line)".Replace("`r", "").Replace("`n", "")
+    Write-Host "Location:" -ForegroundColor Red -NoNewline
+    Write-Host " $($ErrorRecord.InvocationInfo.ScriptName):$($ErrorRecord.InvocationInfo.ScriptLineNumber)"
+    Write-Host "Call Stack:" -ForegroundColor DarkRed
+    $ErrorRecord.ScriptStackTrace -split "`n" | ForEach-Object {
+        Write-Host "    $_" -ForegroundColor DarkRed
+    }
+
+    exit 1
+}
+
+function Write-Info($msg) {
+    Write-Host "[INFO]:" -ForegroundColor Green -NoNewline
+    Write-Host " $msg"
+}
+function Write-Success($msg) {
+    Write-Host "[SUCCESS]:" -ForegroundColor Blue -NoNewline
+    Write-Host " $msg"
+}
+
+
+function Invoke-Conda {
+    param (
+        [Parameter(ValueFromRemainingArguments = $true)]
+        [string[]]$Args
+    )
+
+    $output = & conda install -y -q -c conda-forge @Args 2>&1
+    $exitCode = $LASTEXITCODE
+
+    if ($exitCode -ne 0) {
+        Write-Host "Conda Install $Args Failed" -ForegroundColor Red
+        $errorMessages = @()
+        foreach ($item in $output) {
+            if ($item -is [System.Management.Automation.ErrorRecord]) {
+                $msg = $item.Exception.Message
+                Write-Host "$msg" -ForegroundColor Red
+                $errorMessages += $msg
+            }
+            else {
+                Write-Host $item
+                $errorMessages += $item
+            }
+        }
+        throw [System.Exception]::new(($errorMessages -join "`n"))
+    }
+}
+
+function Invoke-Pip {
+    param (
+        [Parameter(ValueFromRemainingArguments = $true)]
+        [string[]]$Args
+    )
+    
+    $output = & pip install @Args 2>&1
+    $exitCode = $LASTEXITCODE
+    
+    if ($exitCode -ne 0) {
+        $errorMessages = @()
+        Write-Host "Pip Install $Args Failed" -ForegroundColor Red
+        foreach ($item in $output) {
+            if ($item -is [System.Management.Automation.ErrorRecord]) {
+                $msg = $item.Exception.Message
+                Write-Host "$msg" -ForegroundColor Red
+                $errorMessages += $msg
+            }
+            else {
+                Write-Host $item
+                $errorMessages += $item
+            }
+        }
+        throw [System.Exception]::new(($errorMessages -join "`n"))
+    }
+}
+
+function Invoke-Download {
+    param (
+        [Parameter(Mandatory = $true)]
+        [string]$Uri,
+
+        [Parameter()]
+        [string]$OutFile
+    )
+
+    try {
+        $params = @{
+            Uri = $Uri
+        }
+
+        if ($OutFile) {
+            $params["OutFile"] = $OutFile
+        }
+
+        $null = Invoke-WebRequest @params -ErrorAction Stop
+
+    } catch {
+        Write-Host "Failed to download:" -ForegroundColor Red
+        Write-Host "  $Uri"
+        throw
+    }
+}
+
+function Invoke-Unzip {
+    param($ZipPath, $DestPath)
+    Expand-Archive -Path $ZipPath -DestinationPath $DestPath -Force
+    Remove-Item $ZipPath -Force
+}
+
+chcp 65001
+Set-Location $PSScriptRoot
+
+Write-Info "Installing FFmpeg & CMake..."
+Invoke-Conda  ffmpeg cmake
+Write-Success "FFmpeg & CMake Installed"
+
+$PretrainedURL  = ""
+$G2PWURL        = ""
+$UVR5URL        = ""
+$NLTKURL        = ""
+$OpenJTalkURL   = ""
+
+switch ($Source) {
+    "HF" {
+        Write-Info "Download Model From HuggingFace"
+        $PretrainedURL = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/pretrained_models.zip"
+        $G2PWURL       = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/G2PWModel.zip"
+        $UVR5URL       = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/uvr5_weights.zip"
+        $NLTKURL       = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/nltk_data.zip"
+        $OpenJTalkURL  = "https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/open_jtalk_dic_utf_8-1.11.tar.gz"
+    }
+    "HF-Mirror" {
+        Write-Info "Download Model From HuggingFace-Mirror"
+        $PretrainedURL = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/pretrained_models.zip"
+        $G2PWURL       = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/G2PWModel.zip"
+        $UVR5URL       = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/uvr5_weights.zip"
+        $NLTKURL       = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/nltk_data.zip"
+        $OpenJTalkURL  = "https://hf-mirror.com/XXXXRT/GPT-SoVITS-Pretrained/resolve/main/open_jtalk_dic_utf_8-1.11.tar.gz"
+    }
+    "ModelScope" {
+        Write-Info "Download Model From ModelScope"
+        $PretrainedURL = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/pretrained_models.zip"
+        $G2PWURL       = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/G2PWModel.zip"
+        $UVR5URL       = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/uvr5_weights.zip"
+        $NLTKURL       = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/nltk_data.zip"
+        $OpenJTalkURL  = "https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/open_jtalk_dic_utf_8-1.11.tar.gz"
+    }
+}
+
+if (-not (Test-Path "GPT_SoVITS/pretrained_models/sv")) {
+    Write-Info "Downloading Pretrained Models..."
+    Invoke-Download -Uri $PretrainedURL -OutFile "pretrained_models.zip"
+    Invoke-Unzip "pretrained_models.zip" "GPT_SoVITS"
+    Write-Success "Pretrained Models Downloaded"
+} else {
+    Write-Info "Pretrained Model Exists"
+    Write-Info "Skip Downloading Pretrained Models"
+}
+
+
+if (-not (Test-Path "GPT_SoVITS/text/G2PWModel")) {
+    Write-Info "Downloading G2PWModel..."
+    Invoke-Download -Uri $G2PWURL -OutFile "G2PWModel.zip"
+    Invoke-Unzip "G2PWModel.zip" "GPT_SoVITS/text"
+    Write-Success "G2PWModel Downloaded"
+} else {
+    Write-Info "G2PWModel Exists"
+    Write-Info "Skip Downloading G2PWModel"
+}
+
+if ($DownloadUVR5) {
+    if (-not (Test-Path "tools/uvr5/uvr5_weights")) {
+        Write-Info "Downloading UVR5 Models..."
+        Invoke-Download -Uri $UVR5URL -OutFile "uvr5_weights.zip"
+        Invoke-Unzip "uvr5_weights.zip" "tools/uvr5"
+        Write-Success "UVR5 Models Downloaded"
+    } else {
+        Write-Info "UVR5 Models Exists"
+        Write-Info "Skip Downloading UVR5 Models"
+    }
+}
+
+switch ($Device) {
+    "CU128" {
+        Write-Info "Installing PyTorch For CUDA 12.8..."
+        Invoke-Pip torch torchaudio --index-url "https://download.pytorch.org/whl/cu128"
+    }
+    "CU126" {
+        Write-Info "Installing PyTorch For CUDA 12.6..."
+        Invoke-Pip torch torchaudio --index-url "https://download.pytorch.org/whl/cu126"
+    }
+    "CPU" {
+        Write-Info "Installing PyTorch For CPU..."
+        Invoke-Pip torch torchaudio --index-url "https://download.pytorch.org/whl/cpu"
+    }
+}
+Write-Success "PyTorch Installed"
+
+Write-Info "Installing Python Dependencies From requirements.txt..."
+Invoke-Pip -r extra-req.txt --no-deps
+Invoke-Pip -r requirements.txt
+Write-Success "Python Dependencies Installed"
+
+Write-Info "Downloading NLTK Data..."
+Invoke-Download -Uri $NLTKURL -OutFile "nltk_data.zip"
+Invoke-Unzip "nltk_data.zip" (python -c "import sys; print(sys.prefix)").Trim()
+
+Write-Info "Downloading Open JTalk Dict..."
+Invoke-Download -Uri $OpenJTalkURL -OutFile "open_jtalk_dic_utf_8-1.11.tar.gz"
+$target = (python -c "import os, pyopenjtalk; print(os.path.dirname(pyopenjtalk.__file__))").Trim()
+tar -xzf open_jtalk_dic_utf_8-1.11.tar.gz -C $target
+Remove-Item "open_jtalk_dic_utf_8-1.11.tar.gz" -Force
+Write-Success "Open JTalk Dic Downloaded"
+
+Write-Success "Installation Completed"
--- a/install.sh
+++ b/install.sh
@ -48,11 +48,12 @@ run_pip_quiet() {
 }

 run_wget_quiet() {
-    local output
-    output=$(wget --tries=25 --wait=5 --read-timeout=40 --retry-on-http-error=404 "$@" 2>&1) || {
-        echo -e "${ERROR} Wget failed:\n$output"
+    if wget --tries=25 --wait=5 --read-timeout=40 -q --show-progress "$@" 2>&1; then
+        tput cuu1 && tput el
+    else
+        echo -e "${ERROR} Wget failed"
        exit 1
-    }
+    fi
 }

 if ! command -v conda &>/dev/null; then
@ -170,7 +171,16 @@ if ! $USE_HF && ! $USE_HF_MIRROR && ! $USE_MODELSCOPE; then
    exit 1
 fi

-# 安装构建工具
+case "$(uname -m)" in
+x86_64 | amd64) SYSROOT_PKG="sysroot_linux-64>=2.28" ;;
+aarch64 | arm64) SYSROOT_PKG="sysroot_linux-aarch64>=2.28" ;;
+ppc64le) SYSROOT_PKG="sysroot_linux-ppc64le>=2.28" ;;
+*)
+    echo "Unsupported architecture: $(uname -m)"
+    exit 1
+    ;;
+esac
+
 # Install build tools
 echo -e "${INFO}Detected system: $(uname -s) $(uname -r) $(uname -m)"
 if [ "$(uname)" != "Darwin" ]; then
@ -178,10 +188,14 @@ if [ "$(uname)" != "Darwin" ]; then
    if [ "$gcc_major_version" -lt 11 ]; then
        echo -e "${INFO}Installing GCC & G++..."
        run_conda_quiet gcc=11 gxx=11
+        run_conda_quiet "$SYSROOT_PKG"
        echo -e "${SUCCESS}GCC & G++ Installed..."
    else
        echo -e "${INFO}Detected GCC Version: $gcc_major_version"
        echo -e "${INFO}Skip Installing GCC & G++ From Conda-Forge"
+        echo -e "${INFO}Installing libstdcxx-ng From Conda-Forge"
+        run_conda_quiet "libstdcxx-ng>=$gcc_major_version"
+        echo -e "${SUCCESS}libstdcxx-ng=$gcc_major_version Installed..."
    fi
 else
    if ! xcode-select -p &>/dev/null; then
@ -238,10 +252,7 @@ elif [ "$USE_MODELSCOPE" = "true" ]; then
    PYOPENJTALK_URL="https://www.modelscope.cn/models/XXXXRT/GPT-SoVITS-Pretrained/resolve/master/open_jtalk_dic_utf_8-1.11.tar.gz"
 fi

-if find -L "GPT_SoVITS/pretrained_models" -mindepth 1 ! -name '.gitignore' | grep -q .; then
-    echo -e "${INFO}Pretrained Model Exists"
-    echo -e "${INFO}Skip Downloading Pretrained Models"
-else
+if [ ! -d "GPT_SoVITS/pretrained_models/sv" ]; then
    echo -e "${INFO}Downloading Pretrained Models..."
    rm -rf pretrained_models.zip
    run_wget_quiet "$PRETRINED_URL"
@ -249,6 +260,9 @@ else
    unzip -q -o pretrained_models.zip -d GPT_SoVITS
    rm -rf pretrained_models.zip
    echo -e "${SUCCESS}Pretrained Models Downloaded"
+else
+    echo -e "${INFO}Pretrained Model Exists"
+    echo -e "${INFO}Skip Downloading Pretrained Models"
 fi

 if [ ! -d "GPT_SoVITS/text/G2PWModel" ]; then
@ -359,7 +373,7 @@ if [ "$USE_ROCM" = true ] && [ "$IS_WSL" = true ]; then
    location=$(pip show torch | grep Location | awk -F ": " '{print $2}')
    cd "${location}"/torch/lib/ || exit
    rm libhsa-runtime64.so*
-    cp /opt/rocm/lib/libhsa-runtime64.so.1.2 libhsa-runtime64.so
+    cp "$(readlink -f /opt/rocm/lib/libhsa-runtime64.so)" libhsa-runtime64.so
    echo -e "${SUCCESS}ROCm Runtime Lib Updated..."
 fi

--- a/tools/asr/config.py
+++ b/tools/asr/config.py
@ -6,15 +6,10 @@ def check_fw_local_models():
    启动时检查本地是否有 Faster Whisper 模型.
    """
    model_size_list = [
-        "tiny",
-        "tiny.en",
-        "base",
-        "base.en",
-        "small",
-        "small.en",
        "medium",
        "medium.en",
-        "large",
+        "distil-large-v2",
+        "distil-large-v3",
        "large-v1",
        "large-v2",
        "large-v3",
@ -25,11 +20,24 @@ def check_fw_local_models():
    return model_size_list


+def get_models():
+    model_size_list = [
+        "medium",
+        "medium.en",
+        "distil-large-v2",
+        "distil-large-v3",
+        "large-v1",
+        "large-v2",
+        "large-v3",
+    ]
+    return model_size_list
+
+
 asr_dict = {
    "达摩 ASR (中文)": {"lang": ["zh", "yue"], "size": ["large"], "path": "funasr_asr.py", "precision": ["float32"]},
    "Faster Whisper (多语种)": {
        "lang": ["auto", "zh", "en", "ja", "ko", "yue"],
-        "size": check_fw_local_models(),
+        "size": get_models(),
        "path": "fasterwhisper_asr.py",
        "precision": ["float32", "float16", "int8"],
    },
--- a/tools/asr/fasterwhisper_asr.py
+++ b/tools/asr/fasterwhisper_asr.py
@ -1,15 +1,16 @@
 import argparse
 import os
+import time
 import traceback

-os.environ["HF_ENDPOINT"] = "https://hf-mirror.com"
-os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
-
 import torch
 from faster_whisper import WhisperModel
+from huggingface_hub import snapshot_download
+from huggingface_hub.errors import LocalEntryNotFoundError
 from tqdm import tqdm

-from tools.asr.config import check_fw_local_models
+from tools.asr.config import get_models
+from tools.asr.funasr_asr import only_asr
 from tools.my_utils import load_cudnn

 # fmt: off
@ -38,20 +39,54 @@ language_code_list = [
 # fmt: on


-def execute_asr(input_folder, output_folder, model_size, language, precision):
-    if "-local" in model_size:
-        model_size = model_size[:-6]
-        model_path = f"tools/asr/models/faster-whisper-{model_size}"
+def download_model(model_size: str):
+    if "distil" in model_size:
+        repo_id = "Systran/faster-{}-whisper-{}".format(*model_size.split("-", maxsplit=1))
    else:
-        model_path = model_size
+        repo_id = f"Systran/faster-whisper-{model_size}"
+    model_path = f"tools/asr/models/{repo_id.strip('Systran/')}"
+
+    files: list[str] = [
+        "config.json",
+        "model.bin",
+        "tokenizer.json",
+        "vocabulary.txt",
+    ]
+    if model_size == "large-v3" or "distil" in model_size:
+        files.append("preprocessor_config.json")
+        files.append("vocabulary.json")
+
+        files.remove("vocabulary.txt")
+
+    for attempt in range(2):
+        try:
+            snapshot_download(
+                repo_id=repo_id,
+                allow_patterns=files,
+                local_dir=model_path,
+            )
+            break
+        except LocalEntryNotFoundError:
+            if attempt < 1:
+                time.sleep(2)
+            else:
+                print("[ERROR] LocalEntryNotFoundError and no fallback.")
+                traceback.print_exc()
+                exit(1)
+        except Exception as e:
+            print(f"[ERROR] Unexpected error on attempt {attempt + 1}: {e}")
+            traceback.print_exc()
+            exit(1)
+
+    return model_path
+
+
+def execute_asr(input_folder, output_folder, model_path, language, precision):
    if language == "auto":
        language = None  # 不设置语种由模型自动输出概率最高的语种
-    print("loading faster whisper model:", model_size, model_path)
+    print("loading faster whisper model:", model_path, model_path)
    device = "cuda" if torch.cuda.is_available() else "cpu"
-    try:
-        model = WhisperModel(model_path, device=device, compute_type=precision)
-    except:
-        return print(traceback.format_exc())
+    model = WhisperModel(model_path, device=device, compute_type=precision)

    input_file_names = os.listdir(input_folder)
    input_file_names.sort()
@ -73,16 +108,15 @@ def execute_asr(input_folder, output_folder, model_size, language, precision):

            if info.language == "zh":
                print("检测为中文文本, 转 FunASR 处理")
-                if "only_asr" not in globals():
-                    from tools.asr.funasr_asr import only_asr  # 如果用英文就不需要导入下载模型
                text = only_asr(file_path, language=info.language.lower())

            if text == "":
                for segment in segments:
                    text += segment.text
            output.append(f"{file_path}|{output_file_name}|{info.language.upper()}|{text}")
-        except:
-            print(traceback.format_exc())
+        except Exception as e:
+            print(e)
+            traceback.print_exc()

    output_folder = output_folder or "output/asr_opt"
    os.makedirs(output_folder, exist_ok=True)
@ -107,7 +141,7 @@ if __name__ == "__main__":
        "--model_size",
        type=str,
        default="large-v3",
-        choices=check_fw_local_models(),
+        choices=get_models(),
        help="Model Size of Faster Whisper",
    )
    parser.add_argument(
@ -123,10 +157,14 @@ if __name__ == "__main__":
    )

    cmd = parser.parse_args()
+    model_size = cmd.model_size
+    if model_size == "large":
+        model_size = "large-v3"
+    model_path = download_model(model_size)
    output_file_path = execute_asr(
        input_folder=cmd.input_folder,
        output_folder=cmd.output_folder,
-        model_size=cmd.model_size,
+        model_path=model_path,
        language=cmd.language,
        precision=cmd.precision,
    )
--- a/tools/assets.py
+++ b/tools/assets.py
@ -1,6 +1,6 @@
 js = """
 function deleteTheme() {
-    
+
 const params = new URLSearchParams(window.location.search);
 if (params.has('__theme')) {
    params.delete('__theme');
@ -52,13 +52,16 @@ footer * {
 top_html = """
 <div align="center">
    <div style="margin-bottom: 5px; font-size: 15px;">{}</div>
-    <div style="display: flex; gap: 80px; justify-content: center;">
+    <div style="display: flex; gap: 60px; justify-content: center;">
        <a href="https://github.com/RVC-Boss/GPT-SoVITS" target="_blank">
            <img src="https://img.shields.io/badge/GitHub-GPT--SoVITS-blue.svg?style=for-the-badge&logo=github" style="width: auto; height: 30px;">
        </a>
        <a href="https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e" target="_blank">
            <img src="https://img.shields.io/badge/简体中文-阅读文档-blue?style=for-the-badge&logo=googledocs&logoColor=white" style="width: auto; height: 30px;">
        </a>
+        <a href="https://lj1995-gpt-sovits-proplus.hf.space/" target="_blank">
+            <img src="https://img.shields.io/badge/免费在线体验-free_online_demo-yellow.svg?style=for-the-badge&logo=huggingface" style="width: auto; height: 30px;">
+        </a>
        <a href="https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e" target="_blank">
            <img src="https://img.shields.io/badge/English-READ%20DOCS-blue?style=for-the-badge&logo=googledocs&logoColor=white" style="width: auto; height: 30px;">
        </a>
--- a/webui.py
+++ b/webui.py
@ -86,13 +86,10 @@ from config import (
 from tools import my_utils
 from tools.my_utils import check_details, check_for_existance

-# os.environ['PYTORCH_ENABLE_MPS_FALLBACK'] = '1' # 当遇到mps不支持的步骤时使用cpu
-try:
-    import gradio.analytics as analytics
+os.environ["HF_ENDPOINT"] = "https://hf-mirror.com"
+os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"

-    analytics.version_check = lambda: None
-except:
-    ...
+# os.environ['PYTORCH_ENABLE_MPS_FALLBACK'] = '1' # 当遇到mps不支持的步骤时使用cpu
 import gradio as gr

 n_cpu = cpu_count()
Author	SHA1	Message	Date
Oct0pu5	cb7d80ecf8	Merge 2e230d055efc7a37e908a3b28bec8914792397e7 into fdf794e31d1fd6f91c5cb4fbb0396094491a31ac	2025-08-09 02:21:03 +00:00
XXXXRT666	fdf794e31d	Update WSL Rocm (#2561 )	2025-08-02 17:47:15 +08:00
多玩幻灵qwq	0be59c8043	fix: 更正链接 (#2539 )	2025-07-19 00:29:48 +08:00
ChasonJiang	b5a67e6247	修复gpt的loss计算问题 (#2537 ) * 修复gpt的loss计算问题 * fallback tts config	2025-07-18 14:59:59 +08:00
ChasonJiang	b9211657d8	优化TTS_Config的代码逻辑 (#2536 ) * 优化TTS_Config的代码逻辑 * 在载入vits权重之后保存tts_config	2025-07-18 11:54:40 +08:00
XXXXRT666	cefafee32c	Add Distil (#2531 )	2025-07-17 20:28:25 +08:00
RVC-Boss	2d09bbe63a	Update tts_infer.yaml	2025-07-16 15:44:04 +08:00
RVC-Boss	4d8ebf8523	Update TTS.py	2025-07-16 15:43:26 +08:00
jiangsier-xyz	e476b01f30	解决 TTS.py 无法识别真正支持版本 v2Pro、v2ProPlus 的问题 (#2490 ) 同时更新一版默认配置。 Co-authored-by: jiangsier-xyz <jiangsier131@gmail.com>	2025-07-16 15:42:36 +08:00
RVC-Boss	42586e20f7	add RTF performence add RTF performence	2025-07-14 19:01:26 +08:00
RVC-Boss	85035f7ac0	add RTF performence add RTF performence	2025-07-14 18:56:22 +08:00
RVC-Boss	706bec74f8	Update assets.py	2025-07-11 16:11:08 +08:00
XXXXRT666	ec1218893e	Update Badge (#2518 ) * Update README.md * Update README.md * Update Badges * specify ranges	2025-07-11 16:10:07 +08:00
RVC-Boss	fec515dcce	Update Changelog_CN.md	2025-07-10 18:33:18 +08:00
RVC-Boss	426e1a2bb4	提升推理进程优先级	2025-07-10 18:16:45 +08:00
RVC-Boss	4e3c69043c	Update inference_webui.py	2025-07-10 18:16:24 +08:00
RVC-Boss	e63e0901fd	Update assets.py	2025-07-10 18:12:24 +08:00
RVC-Boss	97e37c74d8	Update README.md	2025-07-10 18:06:04 +08:00
RVC-Boss	3a75f5023f	Update README.md	2025-07-10 18:05:03 +08:00
RVC-Boss	0899b7e432	Update README.md	2025-07-10 17:59:49 +08:00
Yixiao Chen	8c579d46dd	Update export_torch_script.py (#2494 ) Avoid dtype inconsistency when exporting	2025-07-02 22:48:28 +08:00
KamioRinn	6df61f58e4	语言分割及格式化优化 (#2488 ) * better LangSegmenter * add version num2str * better version num2str * sync fast infer * sync api * remove duplicate spaces * remove unnecessary code --------- Co-authored-by: RVC-Boss <129054828+RVC-Boss@users.noreply.github.com>	2025-06-27 11:58:41 +08:00
KamioRinn	90ebefa78f	make sure ort providers available (#2489 )	2025-06-27 10:41:52 +08:00
XXXXRT666	4839e82148	Add Windows Install Powershell Scripts (#2487 )	2025-06-27 01:04:18 +08:00
XXXXRT666	37f5abfcb4	Fix Issues with libstdcxx and conda sysroot (#2482 )	2025-06-25 14:52:27 +08:00
Oct0pu5	2e230d055e	Update README.md	2024-09-21 23:02:27 +08:00
Oct0pu5	5726f2a71e	Update README.md	2024-09-21 23:02:11 +08:00
Oct0pu5	8fb80f0646	Update README.md	2024-09-21 23:01:55 +08:00
Oct0pu5	a09a59ae41	Update README.md	2024-09-21 23:01:41 +08:00
Oct0pu5	8de5d0c670	Update README.md	2024-09-21 23:00:28 +08:00