Fix Bugs, Modified Layout (#2434)

Co-authored-by: RVC-Boss <129054828+RVC-Boss@users.noreply.github.com>

GPT_SoVITS/inference_webui.py

@@ -31,9 +31,11 @@ warnings.simplefilter(action="ignore", category=FutureWarning)
 
 version = model_version = os.environ.get("version", "v2")
 
-from config import name2sovits_path,name2gpt_path,change_choices,get_weights_names
+from config import change_choices, get_weights_names, name2gpt_path, name2sovits_path
+
 SoVITS_names, GPT_names = get_weights_names()
 from config import pretrained_sovits_name
+
 path_sovits_v3 = pretrained_sovits_name["v3"]
 path_sovits_v4 = pretrained_sovits_name["v4"]
 is_exist_s2gv3 = os.path.exists(path_sovits_v3)
@@ -108,6 +110,7 @@ from peft import LoraConfig, get_peft_model
 from text import cleaned_text_to_sequence
 from text.cleaner import clean_text
 
+from tools.assets import css, js, top_html
 from tools.i18n.i18n import I18nAuto, scan_language_list
 
 language = os.environ.get("language", "Auto")
@@ -208,8 +211,11 @@ else:
 from process_ckpt import get_sovits_version_from_path_fast, load_sovits_new
 
 v3v4set = {"v3", "v4"}
+
+
 def change_sovits_weights(sovits_path, prompt_language=None, text_language=None):
-    if "！"in sovits_path:sovits_path=name2sovits_path[sovits_path]
+    if "！" 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)
     print(sovits_path, version, model_version, if_lora_v3)
@@ -355,7 +361,8 @@ except:
 
 
 def change_gpt_weights(gpt_path):
-    if "！"in gpt_path:gpt_path=name2gpt_path[gpt_path]
+    if "！" in gpt_path:
+        gpt_path = name2gpt_path[gpt_path]
     global hz, max_sec, t2s_model, config
     hz = 50
     dict_s1 = torch.load(gpt_path, map_location="cpu", weights_only=False)
@@ -383,6 +390,7 @@ import torch
 
 now_dir = os.getcwd()
 
+
 def clean_hifigan_model():
     global hifigan_model
     if hifigan_model:
@@ -392,6 +400,8 @@ def clean_hifigan_model():
             torch.cuda.empty_cache()
         except:
             pass
+
+
 def clean_bigvgan_model():
     global bigvgan_model
     if bigvgan_model:
@@ -401,6 +411,8 @@ def clean_bigvgan_model():
             torch.cuda.empty_cache()
         except:
             pass
+
+
 def clean_sv_cn_model():
     global sv_cn_model
     if sv_cn_model:
@@ -411,6 +423,7 @@ def clean_sv_cn_model():
         except:
             pass
 
+
 def init_bigvgan():
     global bigvgan_model, hifigan_model, sv_cn_model
     from BigVGAN import bigvgan
@@ -429,6 +442,7 @@ def init_bigvgan():
     else:
         bigvgan_model = bigvgan_model.to(device)
 
+
 def init_hifigan():
     global hifigan_model, bigvgan_model, sv_cn_model
     hifigan_model = Generator(
@@ -445,7 +459,9 @@ def init_hifigan():
     hifigan_model.eval()
     hifigan_model.remove_weight_norm()
     state_dict_g = torch.load(
-        "%s/GPT_SoVITS/pretrained_models/gsv-v4-pretrained/vocoder.pth" % (now_dir,), map_location="cpu", weights_only=False
+        "%s/GPT_SoVITS/pretrained_models/gsv-v4-pretrained/vocoder.pth" % (now_dir,),
+        map_location="cpu",
+        weights_only=False,
     )
     print("loading vocoder", hifigan_model.load_state_dict(state_dict_g))
     clean_bigvgan_model()
@@ -455,7 +471,10 @@ def init_hifigan():
     else:
         hifigan_model = hifigan_model.to(device)
 
+
 from sv import SV
+
+
 def init_sv_cn():
     global hifigan_model, bigvgan_model, sv_cn_model
     sv_cn_model = SV(device, is_half)
@@ -472,15 +491,16 @@ if model_version in {"v2Pro","v2ProPlus"}:
     init_sv_cn()
 
 resample_transform_dict = {}
+
+
 def resample(audio_tensor, sr0, sr1, device):
     global resample_transform_dict
     key = "%s-%s-%s" % (sr0, sr1, str(device))
     if key not in resample_transform_dict:
-        resample_transform_dict[key] = torchaudio.transforms.Resample(
+        resample_transform_dict[key] = torchaudio.transforms.Resample(sr0, sr1).to(device)
-            sr0, sr1
-        ).to(device)
     return resample_transform_dict[key](audio_tensor)
 
+
 def get_spepc(hps, filename, dtype, device, is_v2pro=False):
     # audio = load_audio(filename, int(hps.data.sampling_rate))
 
@@ -491,11 +511,13 @@ def get_spepc(hps, filename,dtype,device,is_v2pro=False):
     audio, sr0 = torchaudio.load(filename)
     if sr0 != sr1:
         audio = audio.to(device)
-        if(audio.shape[0]==2):audio=audio.mean(0).unsqueeze(0)
+        if audio.shape[0] == 2:
+            audio = audio.mean(0).unsqueeze(0)
         audio = resample(audio, sr0, sr1, device)
     else:
         audio = audio.to(device)
-        if(audio.shape[0]==2):audio=audio.mean(0).unsqueeze(0)
+        if audio.shape[0] == 2:
+            audio = audio.mean(0).unsqueeze(0)
 
     maxx = audio.abs().max()
     if maxx > 1:
@@ -875,9 +897,13 @@ def get_tts_wav(
                 if is_v2pro:
                     sv_emb = [sv_cn_model.compute_embedding3(audio_tensor)]
             if is_v2pro:
-                audio = vq_model.decode(pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refers, speed=speed,sv_emb=sv_emb)[0][0]
+                audio = vq_model.decode(
+                    pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refers, speed=speed, sv_emb=sv_emb
+                )[0][0]
             else:
-                audio = vq_model.decode(pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refers, speed=speed)[0][0]
+                audio = vq_model.decode(
+                    pred_semantic, torch.LongTensor(phones2).to(device).unsqueeze(0), refers, speed=speed
+                )[0][0]
         else:
             refer, audio_tensor = get_spepc(hps, ref_wav_path, dtype, device)
             phoneme_ids0 = torch.LongTensor(phones1).to(device).unsqueeze(0)
@@ -1076,6 +1102,7 @@ def process_text(texts):
             _text.append(text)
     return _text
 
+
 def html_center(text, label="p"):
     return f"""<div style="text-align: center; margin: 100; padding: 50;">
                 <{label} style="margin: 0; padding: 0;">{text}</{label}>
@@ -1088,11 +1115,13 @@ def html_left(text, label="p"):
                 </div>"""
 
 
-with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False) as app:
+with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False, js=js, css=css) as app:
-    gr.Markdown(
+    gr.HTML(
-        value=i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.")
+        top_html.format(
-        + "<br>"
+            i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.")
             + i18n("如不认可该条款, 则不能使用或引用软件包内任何代码和文件. 详见根目录LICENSE.")
+        ),
+        elem_classes="markdown",
     )
     with gr.Group():
         gr.Markdown(html_center(i18n("模型切换"), "h3"))

GPT_SoVITS/inference_webui_fast.py

@@ -47,6 +47,7 @@ import gradio as gr
 from TTS_infer_pack.text_segmentation_method import get_method
 from TTS_infer_pack.TTS import NO_PROMPT_ERROR, TTS, TTS_Config
 
+from tools.assets import css, js, top_html
 from tools.i18n.i18n import I18nAuto, scan_language_list
 
 language = os.environ.get("language", "Auto")
@@ -98,9 +99,11 @@ cut_method = {
     i18n("按标点符号切"): "cut5",
 }
 
-from config import name2sovits_path,name2gpt_path,change_choices,get_weights_names
+from config import change_choices, get_weights_names, name2gpt_path, name2sovits_path
+
 SoVITS_names, GPT_names = get_weights_names()
 from config import pretrained_sovits_name
+
 path_sovits_v3 = pretrained_sovits_name["v3"]
 path_sovits_v4 = pretrained_sovits_name["v4"]
 is_exist_s2gv3 = os.path.exists(path_sovits_v3)
@@ -111,10 +114,12 @@ tts_config.device = device
 tts_config.is_half = is_half
 tts_config.version = version
 if gpt_path is not None:
-    if "！"in gpt_path:gpt_path=name2gpt_path[gpt_path]
+    if "！" 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:sovits_path=name2sovits_path[sovits_path]
+    if "！" in sovits_path:
+        sovits_path = name2sovits_path[sovits_path]
     tts_config.vits_weights_path = sovits_path
 if cnhubert_base_path is not None:
     tts_config.cnhuhbert_base_path = cnhubert_base_path
@@ -189,6 +194,7 @@ def custom_sort_key(s):
     parts = [int(part) if part.isdigit() else part for part in parts]
     return parts
 
+
 if os.path.exists("./weight.json"):
     pass
 else:
@@ -206,9 +212,13 @@ with open("./weight.json", "r", encoding="utf-8") as file:
         sovits_path = sovits_path[0]
 
 from process_ckpt import get_sovits_version_from_path_fast
+
 v3v4set = {"v3", "v4"}
+
+
 def change_sovits_weights(sovits_path, prompt_language=None, text_language=None):
-    if "！"in sovits_path:sovits_path=name2sovits_path[sovits_path]
+    if "！" 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)
     # print(sovits_path,version, model_version, if_lora_v3)
@@ -273,11 +283,13 @@ def change_sovits_weights(sovits_path, prompt_language=None, text_language=None)
         f.write(json.dumps(data))
 
 
-with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False) as app:
+with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False, js=js, css=css) as app:
-    gr.Markdown(
+    gr.HTML(
-        value=i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.")
+        top_html.format(
-        + "<br>"
+            i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.")
             + i18n("如不认可该条款, 则不能使用或引用软件包内任何代码和文件. 详见根目录LICENSE.")
+        ),
+        elem_classes="markdown",
     )
 
     with gr.Column():

tools/assets.py

@@ -0,0 +1,112 @@
+js = """
+function createGradioAnimation() {
+    
+    const params = new URLSearchParams(window.location.search);
+    if (params.get('__theme') !== 'light') { 
+        params.set('__theme', 'light'); // 仅当 __theme 不是 'light' 时设置为 'light'
+        window.location.search = params.toString(); // 更新 URL，触发页面刷新
+    }
+    
+    var container = document.createElement('div');
+    container.id = 'gradio-animation';
+    container.style.fontSize = '2em';
+    container.style.fontWeight = '500';
+    container.style.textAlign = 'center';
+    container.style.marginBottom = '20px';
+    container.style.fontFamily = '-apple-system, sans-serif, Arial, Calibri';
+
+    var text = 'Welcome to GPT-SoVITS !';
+    for (var i = 0; i < text.length; i++) {
+        (function(i){
+            setTimeout(function(){
+                var letter = document.createElement('span');
+                letter.style.opacity = '0';
+                letter.style.transition = 'opacity 0.5s';
+                letter.innerText = text[i];
+
+                container.appendChild(letter);
+
+                setTimeout(function() {
+                    letter.style.opacity = '1';
+                }, 50);
+            }, i * 250);
+        })(i);
+    }
+    return 'Animation created';
+}
+"""
+
+
+css = """
+/* CSSStyleRule */
+
+.markdown {
+    background-color: lightblue;
+    padding: 10px;
+}
+
+.checkbox_info {
+    color: var(--block-title-text-color) !important;
+    font-size: var(--block-title-text-size) !important;
+    font-weight: var(--block-title-text-weight) !important;
+    height: 22px;
+    margin-bottom: 8px !important;
+}
+
+::selection {
+    background: #ffc078; !important;
+}
+
+#checkbox_train_dpo input[type="checkbox"]{
+    margin-top: 6px;
+}
+
+#checkbox_train_dpo span {
+    margin-top: 6px;
+}
+
+#checkbox_align_train {
+    padding-top: 18px;
+    padding-bottom: 18px;
+}
+
+#checkbox_align_infer input[type="checkbox"] {
+    margin-top: 10px;
+}
+
+#checkbox_align_infer span {
+    margin-top: 10px;
+}
+
+footer {
+    height: 50px !important;           /* 设置页脚高度 */
+    background-color: transparent !important; /* 背景透明 */
+    display: flex;
+    justify-content: center;           /* 居中对齐 */
+    align-items: center;               /* 垂直居中 */
+}
+
+footer * {
+    display: none !important;          /* 隐藏所有子元素 */
+}
+
+"""
+top_html = """
+<div align="center">
+    <div style="margin-bottom: 20px; font-size: 15px;">{}</div>
+    <div style="display: flex; gap: 80px; 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://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 href="https://github.com/RVC-Boss/GPT-SoVITS/blob/main/LICENSE" target="_blank">
+            <img src="https://img.shields.io/badge/LICENSE-MIT-green.svg?style=for-the-badge&logo=opensourceinitiative" style="width: auto; height: 30px;">
+        </a>
+    </div>
+</div>
+"""

webui.py

@@ -60,6 +60,7 @@ import shutil
 import subprocess
 from subprocess import Popen
 
+from tools.assets import css, js, top_html
 from tools.i18n.i18n import I18nAuto, scan_language_list
 
 language = sys.argv[-1] if sys.argv[-1] in scan_language_list() else "Auto"
@@ -1299,14 +1300,13 @@ def sync(text):
     return {"__type__": "update", "value": text}
 
 
-with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False) as app:
+with gr.Blocks(title="GPT-SoVITS WebUI", analytics_enabled=False, js=js, css=css) as app:
-    gr.Markdown(
+    gr.HTML(
-        value=i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.")
+        top_html.format(
+            i18n("本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责.")
             + i18n("如不认可该条款, 则不能使用或引用软件包内任何代码和文件. 详见根目录LICENSE.")
-        + "<br>"
+        ),
-        + i18n("中文教程文档")
+        elem_classes="markdown",
-        + ": "
-        + "https://www.yuque.com/baicaigongchang1145haoyuangong/ib3g1e"
     )
 
     with gr.Tabs():