Merge branch 'main' into ref_audio_selector_tool

GPT_SoVITS/inference_gui.py

@@ -1,3 +1,4 @@
+import os
 import sys
 from PyQt5.QtCore import QEvent
 from PyQt5.QtWidgets import QApplication, QMainWindow, QLabel, QLineEdit, QPushButton, QTextEdit
@@ -7,16 +8,16 @@ import soundfile as sf
 from tools.i18n.i18n import I18nAuto
 i18n = I18nAuto()
 
-from GPT_SoVITS.inference_webui import change_gpt_weights, change_sovits_weights, get_tts_wav
+from inference_webui import gpt_path, sovits_path, change_gpt_weights, change_sovits_weights, get_tts_wav
 
 
 class GPTSoVITSGUI(QMainWindow):
+    GPT_Path = gpt_path
+    SoVITS_Path = sovits_path
+
     def __init__(self):
         super().__init__()
 
-        self.init_ui()
-
-    def init_ui(self):
         self.setWindowTitle('GPT-SoVITS GUI')
         self.setGeometry(800, 450, 950, 850)
 
@@ -71,6 +72,7 @@ class GPTSoVITSGUI(QMainWindow):
         self.GPT_model_label = QLabel("选择GPT模型:")
         self.GPT_model_input = QLineEdit()
         self.GPT_model_input.setPlaceholderText("拖拽或选择文件")
+        self.GPT_model_input.setText(self.GPT_Path)
         self.GPT_model_input.setReadOnly(True)
         self.GPT_model_button = QPushButton("选择GPT模型文件")
         self.GPT_model_button.clicked.connect(self.select_GPT_model)
@@ -78,6 +80,7 @@ class GPTSoVITSGUI(QMainWindow):
         self.SoVITS_model_label = QLabel("选择SoVITS模型:")
         self.SoVITS_model_input = QLineEdit()
         self.SoVITS_model_input.setPlaceholderText("拖拽或选择文件")
+        self.SoVITS_model_input.setText(self.SoVITS_Path)
         self.SoVITS_model_input.setReadOnly(True)
         self.SoVITS_model_button = QPushButton("选择SoVITS模型文件")
         self.SoVITS_model_button.clicked.connect(self.select_SoVITS_model)
@@ -91,25 +94,25 @@ class GPTSoVITSGUI(QMainWindow):
 
         self.ref_text_label = QLabel("参考音频文本:")
         self.ref_text_input = QLineEdit()
-        self.ref_text_input.setPlaceholderText("拖拽或选择文件")
+        self.ref_text_input.setPlaceholderText("直接输入文字或上传文本")
-        self.ref_text_input.setReadOnly(True)
         self.ref_text_button = QPushButton("上传文本")
         self.ref_text_button.clicked.connect(self.upload_ref_text)
 
-        self.language_label = QLabel("参考音频语言:")
+        self.ref_language_label = QLabel("参考音频语言:")
-        self.language_combobox = QComboBox()
+        self.ref_language_combobox = QComboBox()
-        self.language_combobox.addItems(["中文", "英文", "日文"])
+        self.ref_language_combobox.addItems(["中文", "英文", "日文", "中英混合", "日英混合", "多语种混合"])
+        self.ref_language_combobox.setCurrentText("多语种混合")
 
         self.target_text_label = QLabel("合成目标文本:")
         self.target_text_input = QLineEdit()
-        self.target_text_input.setPlaceholderText("拖拽或选择文件")
+        self.target_text_input.setPlaceholderText("直接输入文字或上传文本")
-        self.target_text_input.setReadOnly(True)
         self.target_text_button = QPushButton("上传文本")
         self.target_text_button.clicked.connect(self.upload_target_text)
 
-        self.language_label_02 = QLabel("合成音频语言:")
+        self.target_language_label = QLabel("合成音频语言:")
-        self.language_combobox_02 = QComboBox()
+        self.target_language_combobox = QComboBox()
-        self.language_combobox_02.addItems(["中文", "英文", "日文"])
+        self.target_language_combobox.addItems(["中文", "英文", "日文", "中英混合", "日英混合", "多语种混合"])
+        self.target_language_combobox.setCurrentText("多语种混合")
 
         self.output_label = QLabel("输出音频路径:")
         self.output_input = QLineEdit()
@@ -140,10 +143,8 @@ class GPTSoVITSGUI(QMainWindow):
 
         main_layout = QVBoxLayout()
 
-        input_layout = QGridLayout()
+        input_layout = QGridLayout(self)
-        input_layout.setSpacing(10) 
+        input_layout.setSpacing(10)
-
-        self.setLayout(input_layout)
 
         input_layout.addWidget(license_label, 0, 0, 1, 3)
 
@@ -159,22 +160,22 @@ class GPTSoVITSGUI(QMainWindow):
         input_layout.addWidget(self.ref_audio_input, 6, 0, 1, 2)
         input_layout.addWidget(self.ref_audio_button, 6, 2)
 
-        input_layout.addWidget(self.language_label, 7, 0)
+        input_layout.addWidget(self.ref_language_label, 7, 0)
-        input_layout.addWidget(self.language_combobox, 8, 0, 1, 1)
+        input_layout.addWidget(self.ref_language_combobox, 8, 0, 1, 1)
         input_layout.addWidget(self.ref_text_label, 9, 0)
         input_layout.addWidget(self.ref_text_input, 10, 0, 1, 2)
         input_layout.addWidget(self.ref_text_button, 10, 2)
 
-        input_layout.addWidget(self.language_label_02, 11, 0)
+        input_layout.addWidget(self.target_language_label, 11, 0)
-        input_layout.addWidget(self.language_combobox_02, 12, 0, 1, 1)
+        input_layout.addWidget(self.target_language_combobox, 12, 0, 1, 1)
         input_layout.addWidget(self.target_text_label, 13, 0)
         input_layout.addWidget(self.target_text_input, 14, 0, 1, 2)
         input_layout.addWidget(self.target_text_button, 14, 2)
-        
+
         input_layout.addWidget(self.output_label, 15, 0)
         input_layout.addWidget(self.output_input, 16, 0, 1, 2)
         input_layout.addWidget(self.output_button, 16, 2)
-        
+
         main_layout.addLayout(input_layout)
 
         output_layout = QVBoxLayout()
@@ -198,10 +199,8 @@ class GPTSoVITSGUI(QMainWindow):
     def dropEvent(self, event):
         if event.mimeData().hasUrls():
             file_paths = [url.toLocalFile() for url in event.mimeData().urls()]
-          
             if len(file_paths) == 1:
                 self.update_ref_audio(file_paths[0])
-                self.update_input_paths(self.ref_audio_input, file_paths[0])
             else:
                 self.update_ref_audio(", ".join(file_paths))
 
@@ -211,23 +210,13 @@ class GPTSoVITSGUI(QMainWindow):
             widget.installEventFilter(self)
 
     def eventFilter(self, obj, event):
-        if event.type() == QEvent.DragEnter:
+        if event.type() in (QEvent.DragEnter, QEvent.Drop):
             mime_data = event.mimeData()
             if mime_data.hasUrls():
                 event.acceptProposedAction()
-              
-        elif event.type() == QEvent.Drop:
-            mime_data = event.mimeData()
-            if mime_data.hasUrls():
-                file_paths = [url.toLocalFile() for url in mime_data.urls()]
-                if len(file_paths) == 1:
-                    self.update_input_paths(obj, file_paths[0])
-                else:
-                    self.update_input_paths(obj, ", ".join(file_paths))
-                event.acceptProposedAction()
 
         return super().eventFilter(obj, event)
-    
+
     def select_GPT_model(self):
         file_path, _ = QFileDialog.getOpenFileName(self, "选择GPT模型文件", "", "GPT Files (*.ckpt)")
         if file_path:
@@ -239,24 +228,9 @@ class GPTSoVITSGUI(QMainWindow):
             self.SoVITS_model_input.setText(file_path)
 
     def select_ref_audio(self):
-        options = QFileDialog.Options()
+        file_path, _ = QFileDialog.getOpenFileName(self, "选择参考音频文件", "", "Audio Files (*.wav *.mp3)")
-        options |= QFileDialog.DontUseNativeDialog
+        if file_path:
-        options |= QFileDialog.ShowDirsOnly
+            self.update_ref_audio(file_path)
-
-        file_dialog = QFileDialog()
-        file_dialog.setOptions(options)
-
-        file_dialog.setFileMode(QFileDialog.AnyFile)
-        file_dialog.setNameFilter("Audio Files (*.wav *.mp3)")
-
-        if file_dialog.exec_():
-            file_paths = file_dialog.selectedFiles()
-          
-            if len(file_paths) == 1:
-                self.update_ref_audio(file_paths[0])
-                self.update_input_paths(self.ref_audio_input, file_paths[0])
-            else:
-                self.update_ref_audio(", ".join(file_paths))
 
     def upload_ref_text(self):
         file_path, _ = QFileDialog.getOpenFileName(self, "选择文本文件", "", "Text Files (*.txt)")
@@ -264,7 +238,6 @@ class GPTSoVITSGUI(QMainWindow):
             with open(file_path, 'r', encoding='utf-8') as file:
                 content = file.read()
                 self.ref_text_input.setText(content)
-                self.update_input_paths(self.ref_text_input, file_path)
 
     def upload_target_text(self):
         file_path, _ = QFileDialog.getOpenFileName(self, "选择文本文件", "", "Text Files (*.txt)")
@@ -272,7 +245,6 @@ class GPTSoVITSGUI(QMainWindow):
             with open(file_path, 'r', encoding='utf-8') as file:
                 content = file.read()
                 self.target_text_input.setText(content)
-                self.update_input_paths(self.target_text_input, file_path)
 
     def select_output_path(self):
         options = QFileDialog.Options()
@@ -290,9 +262,6 @@ class GPTSoVITSGUI(QMainWindow):
     def update_ref_audio(self, file_path):
         self.ref_audio_input.setText(file_path)
 
-    def update_input_paths(self, input_box, file_path):
-        input_box.setText(file_path)
-
     def clear_output(self):
         self.output_text.clear()
 
@@ -300,23 +269,27 @@ class GPTSoVITSGUI(QMainWindow):
         GPT_model_path = self.GPT_model_input.text()
         SoVITS_model_path = self.SoVITS_model_input.text()
         ref_audio_path = self.ref_audio_input.text()
-        language_combobox = self.language_combobox.currentText()
+        language_combobox = self.ref_language_combobox.currentText()
         language_combobox = i18n(language_combobox)
         ref_text = self.ref_text_input.text()
-        language_combobox_02 = self.language_combobox_02.currentText()
+        target_language_combobox = self.target_language_combobox.currentText()
-        language_combobox_02 = i18n(language_combobox_02)
+        target_language_combobox = i18n(target_language_combobox)
         target_text = self.target_text_input.text()
         output_path = self.output_input.text()
 
-        change_gpt_weights(gpt_path=GPT_model_path)
+        if GPT_model_path != self.GPT_Path:
-        change_sovits_weights(sovits_path=SoVITS_model_path)
+            change_gpt_weights(gpt_path=GPT_model_path)
+            self.GPT_Path = GPT_model_path
+        if SoVITS_model_path != self.SoVITS_Path:
+            change_sovits_weights(sovits_path=SoVITS_model_path)
+            self.SoVITS_Path = SoVITS_model_path
 
         synthesis_result = get_tts_wav(ref_wav_path=ref_audio_path, 
                                        prompt_text=ref_text, 
                                        prompt_language=language_combobox, 
                                        text=target_text, 
-                                       text_language=language_combobox_02)
+                                       text_language=target_language_combobox)
-        
+
         result_list = list(synthesis_result)
 
         if result_list:
@@ -329,12 +302,9 @@ class GPTSoVITSGUI(QMainWindow):
         self.status_bar.showMessage("合成完成！输出路径：" + output_wav_path, 5000)
         self.output_text.append("处理结果：\n" + result)
 
-def main():
+
+if __name__ == '__main__':
     app = QApplication(sys.argv)
     mainWin = GPTSoVITSGUI()
     mainWin.show()
-    sys.exit(app.exec_())
+    sys.exit(app.exec_())
-
-
-if __name__ == '__main__':
-    main()

GPT_SoVITS/inference_webui.py

@@ -50,6 +50,7 @@ is_share = eval(is_share)
 if "_CUDA_VISIBLE_DEVICES" in os.environ:
     os.environ["CUDA_VISIBLE_DEVICES"] = os.environ["_CUDA_VISIBLE_DEVICES"]
 is_half = eval(os.environ.get("is_half", "True")) and torch.cuda.is_available()
+punctuation = set(['!', '?', '…', ',', '.', '-'," "])
 import gradio as gr
 from transformers import AutoModelForMaskedLM, AutoTokenizer
 import numpy as np
@@ -322,6 +323,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         if (prompt_text[-1] not in splits): prompt_text += "。" if prompt_language != "en" else "."
         print(i18n("实际输入的参考文本:"), prompt_text)
     text = text.strip("\n")
+    text = replace_consecutive_punctuation(text)
     if (text[0] not in splits and len(get_first(text)) < 4): text = "。" + text if text_language != "en" else "." + text
     
     print(i18n("实际输入的目标文本:"), text)
@@ -366,6 +368,7 @@ def get_tts_wav(ref_wav_path, prompt_text, prompt_language, text, text_language,
         text = text.replace("\n\n", "\n")
     print(i18n("实际输入的目标文本(切句后):"), text)
     texts = text.split("\n")
+    texts = process_text(texts)
     texts = merge_short_text_in_array(texts, 5)
     audio_opt = []
     if not ref_free:
@@ -463,6 +466,7 @@ def cut1(inp):
             opts.append("".join(inps[split_idx[idx]: split_idx[idx + 1]]))
     else:
         opts = [inp]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
     return "\n".join(opts)
 
 
@@ -487,17 +491,21 @@ def cut2(inp):
     if len(opts) > 1 and len(opts[-1]) < 50:  ##如果最后一个太短了，和前一个合一起
         opts[-2] = opts[-2] + opts[-1]
         opts = opts[:-1]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
     return "\n".join(opts)
 
 
 def cut3(inp):
     inp = inp.strip("\n")
-    return "\n".join(["%s" % item for item in inp.strip("。").split("。")])
+    opts = ["%s" % item for item in inp.strip("。").split("。")]
-
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
+    return  "\n".join(opts)
 
 def cut4(inp):
     inp = inp.strip("\n")
-    return "\n".join(["%s" % item for item in inp.strip(".").split(".")])
+    opts = ["%s" % item for item in inp.strip(".").split(".")]
+    opts = [item for item in opts if not set(item).issubset(punctuation)]
+    return "\n".join(opts)
 
 
 # contributed by https://github.com/AI-Hobbyist/GPT-SoVITS/blob/main/GPT_SoVITS/inference_webui.py
@@ -511,8 +519,8 @@ def cut5(inp):
     # 在句子不存在符号或句尾无符号的时候保证文本完整
     if len(items)%2 == 1:
         mergeitems.append(items[-1])
-    opt = "\n".join(mergeitems)
+    opt = [item for item in mergeitems if not set(item).issubset(punctuation)]
-    return opt
+    return "\n".join(opt)
 
 
 def custom_sort_key(s):
@@ -522,6 +530,24 @@ def custom_sort_key(s):
     parts = [int(part) if part.isdigit() else part for part in parts]
     return parts
 
+def process_text(texts):
+    _text=[]
+    if all(text in [None, " ", "\n",""] for text in texts):
+        raise ValueError(i18n("请输入有效文本"))
+    for text in texts:
+        if text in  [None, " ", ""]:
+            pass
+        else:
+            _text.append(text)
+    return _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 change_choices():
     SoVITS_names, GPT_names = get_weights_names()
@@ -613,10 +639,12 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             button5.click(cut5, [text_inp], [text_opt])
         gr.Markdown(value=i18n("后续将支持转音素、手工修改音素、语音合成分步执行。"))
 
-app.queue(concurrency_count=511, max_size=1022).launch(
+
-    server_name="0.0.0.0",
+if __name__ == '__main__':
-    inbrowser=True,
+    app.queue(concurrency_count=511, max_size=1022).launch(
-    share=is_share,
+        server_name="0.0.0.0",
-    server_port=infer_ttswebui,
+        inbrowser=True,
-    quiet=True,
+        share=is_share,
-)
+        server_port=infer_ttswebui,
+        quiet=True,
+    )

GPT_SoVITS/s2_train.py

@@ -305,7 +305,7 @@ def train_and_evaluate(
         y_lengths,
         text,
         text_lengths,
-    ) in tqdm(enumerate(train_loader)):
+    ) in enumerate(tqdm(train_loader)):
         if torch.cuda.is_available():
             spec, spec_lengths = spec.cuda(rank, non_blocking=True), spec_lengths.cuda(
                 rank, non_blocking=True

docs/cn/Changelog_CN.md

@@ -169,6 +169,21 @@
 
 6-nan自动转fp32阶段的hubert提取bug修复
 
+### 20240610
+
+小问题修复：
+
+1-完善纯标点、多标点文本输入的判断逻辑 https://github.com/RVC-Boss/GPT-SoVITS/pull/1168 https://github.com/RVC-Boss/GPT-SoVITS/pull/1169
+
+2-uvr5中的mdxnet去混响cmd格式修复，兼容路径带空格  [#501a74a](https://github.com/RVC-Boss/GPT-SoVITS/commit/501a74ae96789a26b48932babed5eb4e9483a232)
+
+3-s2训练进度条逻辑修复 https://github.com/RVC-Boss/GPT-SoVITS/pull/1159
+
+大问题修复：
+
+4-修复了webui的GPT中文微调没读到bert导致和推理不一致，训练太多可能效果还会变差的问题。如果大量数据微调的建议重新微调模型得到质量优化 [#99f09c8](https://github.com/RVC-Boss/GPT-SoVITS/commit/99f09c8bdc155c1f4272b511940717705509582a)
+
+
 todolist：
 
 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
@@ -177,3 +192,5 @@ todolist：
 2-正在尝试解决低音质参考音频导致音质较差的问题，v2再试试如果能解决就发了，节点暂定高考后吧
 
 
+
+

tools/uvr5/mdxnet.py

@@ -220,7 +220,7 @@ class Predictor:
             opt_path_other = path_other[:-4] + ".%s" % format
             if os.path.exists(path_vocal):
                 os.system(
-                    "ffmpeg -i %s -vn %s -q:a 2 -y" % (path_vocal, opt_path_vocal)
+                    "ffmpeg -i '%s' -vn '%s' -q:a 2 -y" % (path_vocal, opt_path_vocal)
                 )
                 if os.path.exists(opt_path_vocal):
                     try:
@@ -229,7 +229,7 @@ class Predictor:
                         pass
             if os.path.exists(path_other):
                 os.system(
-                    "ffmpeg -i %s -vn %s -q:a 2 -y" % (path_other, opt_path_other)
+                    "ffmpeg -i '%s' -vn '%s' -q:a 2 -y" % (path_other, opt_path_other)
                 )
                 if os.path.exists(opt_path_other):
                     try:

webui.py

@@ -85,7 +85,7 @@ if if_gpu_ok and len(gpu_infos) > 0:
 else:
     gpu_info = ("%s\t%s" % ("0", "CPU"))
     gpu_infos.append("%s\t%s" % ("0", "CPU"))
-    default_batch_size = psutil.virtual_memory().total/ 1024 / 1024 / 1024 / 2
+    default_batch_size = int(psutil.virtual_memory().total/ 1024 / 1024 / 1024 / 2)
 gpus = "-".join([i[0] for i in gpu_infos])
 
 pretrained_sovits_name="GPT_SoVITS/pretrained_models/s2G488k.pth"