Merge b2d725ed9a2d50465f7882ecdd06715d66ef66bd into d6b78c927acf89dfc12acb1bc561c1c963543ecf

* fix configs error

* fix configs error

---------

Co-authored-by: wangzeyuan <wangzeyuan@agora.io>
Co-authored-by: wangzeyuan <wangzeyuan@shengwang.cn>

GPT_SoVITS/inference_webui.py

@@ -214,7 +214,7 @@ v3v4set = {"v3", "v4"}
 
 
 def change_sovits_weights(sovits_path, prompt_language=None, text_language=None):
-    if "！" in sovits_path:
+    if "！" in sovits_path or "!" in sovits_path:
         sovits_path = name2sovits_path[sovits_path]
     global vq_model, hps, version, model_version, dict_language, if_lora_v3
     version, model_version, if_lora_v3 = get_sovits_version_from_path_fast(sovits_path)
@@ -361,7 +361,7 @@ except:
 
 
 def change_gpt_weights(gpt_path):
-    if "！" in gpt_path:
+    if "！" in gpt_path or "!" in gpt_path:
         gpt_path = name2gpt_path[gpt_path]
     global hz, max_sec, t2s_model, config
     hz = 50

GPT_SoVITS/inference_webui_fast.py

@@ -114,11 +114,11 @@ tts_config.device = device
 tts_config.is_half = is_half
 tts_config.version = version
 if gpt_path is not None:
-    if "！" in gpt_path:
+    if "！" in gpt_path or "!" in gpt_path:
         gpt_path = name2gpt_path[gpt_path]
     tts_config.t2s_weights_path = gpt_path
 if sovits_path is not None:
-    if "！" in sovits_path:
+    if "！" in sovits_path or "!" in sovits_path:
         sovits_path = name2sovits_path[sovits_path]
     tts_config.vits_weights_path = sovits_path
 if cnhubert_base_path is not None:
@@ -217,7 +217,7 @@ v3v4set = {"v3", "v4"}
 
 
 def change_sovits_weights(sovits_path, prompt_language=None, text_language=None):
-    if "！" in sovits_path:
+    if "！" in sovits_path or "!" in sovits_path:
         sovits_path = name2sovits_path[sovits_path]
     global version, model_version, dict_language, if_lora_v3
     version, model_version, if_lora_v3 = get_sovits_version_from_path_fast(sovits_path)
@@ -283,6 +283,12 @@ def change_sovits_weights(sovits_path, prompt_language=None, text_language=None)
         f.write(json.dumps(data))
 
 
+def change_gpt_weights(gpt_path):
+    if "！" in gpt_path or "!" in gpt_path:
+        gpt_path = name2gpt_path[gpt_path]
+    tts_pipeline.init_t2s_weights(gpt_path)
+
+
 with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False, js=js, css=css) as app:
     gr.HTML(
         top_html.format(
@@ -457,7 +463,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False, js=js, css=css
                 inference_button,
             ],
         )  #
-        GPT_dropdown.change(tts_pipeline.init_t2s_weights, [GPT_dropdown], [])
+        GPT_dropdown.change(change_gpt_weights, [GPT_dropdown], [])
 
     with gr.Group():
         gr.Markdown(

GPT_SoVITS/process_ckpt.py

@@ -127,7 +127,7 @@ def get_sovits_version_from_path_fast(sovits_path):
 def load_sovits_new(sovits_path):
     f = open(sovits_path, "rb")
     meta = f.read(2)
-    if meta != "PK":
+    if meta != b"PK":
         data = b"PK" + f.read()
         bio = BytesIO()
         bio.write(data)

api.py

@@ -313,9 +313,11 @@ def get_sovits_weights(sovits_path):
 
     if model_version == "v3":
         hps.model.version = "v3"
+    if model_version == "v4":
+        hps.model.version = "v4"
 
     model_params_dict = vars(hps.model)
-    if model_version != "v3":
+    if model_version != "v3" and model_version != "v4":
         vq_model = SynthesizerTrn(
             hps.data.filter_length // 2 + 1,
             hps.train.segment_size // hps.data.hop_length,
@@ -759,7 +761,7 @@ def get_tts_wav(
         prompt_semantic = codes[0, 0]
         prompt = prompt_semantic.unsqueeze(0).to(device)
 
-        if version != "v3":
+        if version != "v3" and version != "v4":
             refers = []
             if inp_refs:
                 for path in inp_refs:
@@ -811,7 +813,7 @@ def get_tts_wav(
             pred_semantic = pred_semantic[:, -idx:].unsqueeze(0)
         t3 = ttime()
 
-        if version != "v3":
+        if version != "v3" and version != "v4":
             audio = (
                 vq_model.decode(pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refers, speed=speed)
                 .detach()