Merge dffefeab8422bb3a452b31f5b898dc19fd0c87f1 into 9da7e17efe05041e31d3c3f42c8730ae890397f2

- Address parameter assignment issue for DefaultRefer class
- Standarlize parameter passing in initialization section

GPT_SoVITS/inference_webui_fast.py

@@ -227,8 +227,8 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
     global version, model_version, dict_language,if_lora_v3
     version, model_version, if_lora_v3=get_sovits_version_from_path_fast(sovits_path)
     # print(sovits_path,version, model_version, if_lora_v3)
-    if if_lora_v3==True and is_exist_s2gv3==False:#
-        info= "GPT_SoVITS/pretrained_models/s2Gv3.pth" + i18n("SoVITS V3 底模缺失，无法加载相应 LoRA 权重")
+    if if_lora_v3 and not os.path.exists(path_sovits_v3):
+        info= path_sovits_v3 + i18n("SoVITS V3 底模缺失，无法加载相应 LoRA 权重")
         gr.Warning(info)
         raise FileExistsError(info)
     dict_language = dict_language_v1 if version =='v1' else dict_language_v2

GPT_SoVITS/text/g2pw/onnx_api.py

@@ -58,7 +58,7 @@ def download_and_decompress(model_dir: str='G2PWModel/'):
         extract_dir = os.path.join(parent_directory,"G2PWModel_1.1")
         extract_dir_new = os.path.join(parent_directory,"G2PWModel")
         print("Downloading g2pw model...")
-        modelscope_url = "https://paddlespeech.cdn.bcebos.com/Parakeet/released_models/g2p/G2PWModel_1.1.zip"
+        modelscope_url = "https://www.modelscope.cn/models/kamiorinn/g2pw/resolve/master/G2PWModel_1.1.zip"#"https://paddlespeech.cdn.bcebos.com/Parakeet/released_models/g2p/G2PWModel_1.1.zip"
         with requests.get(modelscope_url, stream=True) as r:
             r.raise_for_status()
             with open(zip_dir, 'wb') as f:

api.py

@@ -176,9 +176,9 @@ import subprocess
 
 class DefaultRefer:
     def __init__(self, path, text, language):
-        self.path = args.default_refer_path
-        self.text = args.default_refer_text
-        self.language = args.default_refer_language
+        self.path = path
+        self.text = text
+        self.language = language
 
     def is_ready(self) -> bool:
         return is_full(self.path, self.text, self.language)
@@ -934,15 +934,23 @@ parser.add_argument("-b", "--bert_path", type=str, default=g_config.bert_path, h
 args = parser.parse_args()
 sovits_path = args.sovits_path
 gpt_path = args.gpt_path
+default_refer_path = args.default_refer_path
+default_refer_text = args.default_refer_text
+default_refer_language = args.default_refer_language
 device = args.device
 port = args.port
 host = args.bind_addr
+full_precision = args.full_precision
+half_precision = args.half_precision
+stream_mode = args.stream_mode
+media_type = args.media_type
+sub_type = args.sub_type
+default_cut_punc = args.cut_punc
 cnhubert_base_path = args.hubert_path
 bert_path = args.bert_path
-default_cut_punc = args.cut_punc
 
 # 应用参数配置
-default_refer = DefaultRefer(args.default_refer_path, args.default_refer_text, args.default_refer_language)
+default_refer = DefaultRefer(default_refer_path, default_refer_text, default_refer_language)
 
 # 模型路径检查
 if sovits_path == "":
@@ -963,24 +971,24 @@ else:
 
 # 获取半精度
 is_half = g_config.is_half
-if args.full_precision:
+if full_precision:
     is_half = False
-if args.half_precision:
+if half_precision:
     is_half = True
-if args.full_precision and args.half_precision:
+if full_precision and half_precision:
     is_half = g_config.is_half  # 炒饭fallback
 logger.info(f"半精: {is_half}")
 
 # 流式返回模式
-if args.stream_mode.lower() in ["normal","n"]:
+if stream_mode.lower() in ["normal","n"]:
     stream_mode = "normal"
     logger.info("流式返回已开启")
 else:
     stream_mode = "close"
 
 # 音频编码格式
-if args.media_type.lower() in ["aac","ogg"]:
-    media_type = args.media_type.lower()
+if media_type.lower() in ["aac","ogg"]:
+    media_type = media_type.lower()
 elif stream_mode == "close":
     media_type = "wav"
 else:
@@ -988,7 +996,7 @@ else:
 logger.info(f"编码格式: {media_type}")
 
 # 音频数据类型
-if args.sub_type.lower() == 'int32':
+if sub_type.lower() == 'int32':
     is_int32 = True
     logger.info(f"数据类型: int32")
 else:

docs/cn/Changelog_CN.md

@@ -286,3 +286,17 @@ https://github.com/RVC-Boss/GPT-SoVITS/pull/2112 https://github.com/RVC-Boss/GPT
 修复短文本语种选择出错 https://github.com/RVC-Boss/GPT-SoVITS/pull/2122
 
 修复v3sovits未传参以支持调节语速
+
+### 202503
+
+修复一批由依赖的库版本不对导致的问题https://github.com/RVC-Boss/GPT-SoVITS/commit/6c468583c5566e5fbb4fb805e4cc89c403e997b8
+
+修复模型加载异步逻辑https://github.com/RVC-Boss/GPT-SoVITS/commit/03b662a769946b7a6a8569a354860e8eeeb743aa
+
+修复其他若干bug
+
+重点更新：
+
+1-v3支持并行推理 https://github.com/RVC-Boss/GPT-SoVITS/commit/03b662a769946b7a6a8569a354860e8eeeb743aa
+
+2-整合包修复onnxruntime GPU推理的支持，影响：（1）g2pw有个onnx模型原先是CPU推理现在用GPU，显著降低推理的CPU瓶颈 （2）foxjoy去混响模型现在可使用GPU推理