support pre-built local plugin

Author: andantei

README.md

@@ -10,6 +10,16 @@ git clone https://github.com/tuneflow/so-vits-svc-plugin.git
 
 ## Installation
 
+### (Option 1) Download Pre-Built Release
+
+You can also download a prebuilt package with all dependencies included, from the below links:
+
+[Download for Windows](https://plugin-dist.tuneflow.com/plugins/binary/svc_local/win-x64-1.0.0.zip)
+
+[Download for macOS (Coming Soon)](#)
+
+### (Option 2) Build from Source
+
 It is recommended to install through python virtual environments, so that these dependencies won't conflict with your existing pip packages.
 
 ```bash
@@ -25,7 +35,7 @@ If you want the plugin to load models on start up, you can optionally place your
 The folder structure would look like this:
 
 ```
--- so-vits-svc-plugin
+-- <root folder of the project>
     -- ......
     -- checkpoints
         -- G_XXX.pth
@@ -35,13 +45,19 @@ The folder structure would look like this:
 
 ## Run the Plugin
 
+### If You Downloaded Pre-built Release
+
+Unzip the app, and run the `local_app.exe` from the extracted folder.
+
+### If You Built from Source
+
 Once you installed the dependencies and prepared the models, you can start running the plugin using:
 
 ```bash
-python local_plugin.py
+python local_app.py
 ```
 
-You should see something like this in your console log:
+Once started, you should be able to see something like this:
 
 ```bash
 ============= Plugin Info =============
@@ -57,7 +73,7 @@ Next, start TuneFlow Desktop, if you don't have it already, download from the ho
 
 Create an empty song, or open an existing song.
 
-Once the project is loaded, switch to the TuneFlow Plugin Library, at the top right corner, click on the "**Load a remotely hosted plugin**" button. And paste `http://127.0.0.1:8000/plugins/singing-voice-clone-local` into the address input. If everything is setup correctly our plugin should load successfully and show up in the plugin inventory.
+Once the project is loaded, switch to the TuneFlow Plugin Library, at the top right corner, click on the "Load a remotely hosted plugin" button. And paste http://127.0.0.1:8000/plugins/singing-voice-clone-local into the address input. If everything is setup correctly our plugin should load successfully and show up in the plugin inventory.
 
 ![Load local plugin](./images/load_plugin_en.jpg)
 

README.zh.md

@@ -8,7 +8,19 @@
 git clone https://github.com/tuneflow/so-vits-svc-plugin.git
 ```
 
-## 安装依赖
+## 安装
+
+### （选项1）使用整合好的压缩包
+
+首先下载安装包：
+
+[Windows下载](https://plugin-dist.tuneflow.com/plugins/binary/svc_local/win-x64-1.0.0.zip)
+
+[macOS下载(即将推出)](#)
+
+随后解压下载好的安装包，其中的`local_app.exe`即为插件运行程序。
+
+### （选项2）从源代码安装
 
 注意：安装 python 依赖包时推荐使用 virtualenv，这样可以将插件所需依赖与系统其他依赖分离开来。
 
@@ -23,26 +35,36 @@ pip install -r requirements.txt
 如果你想每次运行插件时都自动加载模型，你可以把模型和配置文件放到`checkpoints`目录下，这样整个目录结构看起来是这样的：
 
 ```
--- so-vits-svc-plugin
+-- <项目根目录>
     -- ......
     -- checkpoints
         -- G_XXX.pth
         -- config.json
     -- ......
 ```
 
+该方法对于两种安装方式都适用。
+
 ## 训练自己的模型
 
 参考 [炼丹百科全书](https://docs.qq.com/doc/DUWdxS1ZaV29vZnlV)，有非常易用的网页版训练界面。
 
 ## 运行插件
 
+### 如果你下载了打包好的压缩包
+
+解压下载好的压缩包，运行其中的`local_app.exe`文件。
+
+### 如果你是从源文件直接构建
+
 当 python 依赖和模型文件准备完毕后，我们可以开始跑起来插件了。用以下命令运行插件:
 
 ```bash
-python local_plugin.py
+python local_app.py
 ```
 
+### 在 TuneFlow 中加载插件
+
 插件正常运行的情况下，你可以看到类似下面的控制台输出：
 
 ```bash
@@ -59,7 +81,7 @@ Plugin Description: Sing a vocal clip with a new voice
 
 打开桌面版后，我们可以创建一首空白曲目，或者打开一首已有的曲子。
 
-曲目加载完成后，在界面右侧切换到 TuneFlow 插件库。在插件库面板的右上角点击"**加载远程插件**"按钮，并在地址框中输入`http://127.0.0.1:8000/plugins/singing-voice-clone-local`。如果所有东西配置正确的话，你应该可以看到插件被加载到了插件库中。
+曲目加载完成后，在界面右侧切换到 TuneFlow 插件库。在插件库面板的右上角点击"加载远程插件"按钮，并在地址框中输入http://127.0.0.1:8000/plugins/singing-voice-clone-local。如果所有东西配置正确的话，你应该可以看到插件被加载到了插件库中。
 
 ![加载本地插件](./images/load_plugin_zh.jpg)
 

bundle_local.json

@@ -17,7 +17,7 @@
       },
       "version": "1.0.0",
       "supportedPlatforms": ["desktop"],
-      "minRequiredDesktopVersion": "1.9.1",
+      "minRequiredDesktopVersion": "1.9.3",
       "options": {
         "allowReset": true,
         "allowManualApplyAdjust": true
@@ -74,11 +74,8 @@
       "categories": ["generate"],
       "localPlugin": {
         "binaryUrl": {
-          "mac": {
-            "x64": "https://s.tuneflow.com/plugins/binary/singing-voice-clone-local/mac-x64"
-          },
           "win": {
-            "x64": "https://s.tuneflow.com/plugins/binary/singing-voice-clone-local/win-x64"
+            "x64": "https://plugin-dist.tuneflow.com/plugins/binary/svc_local/win-x64-1.0.0.zip"
           }
         },
         "servingPath": "/plugins/singing-voice-clone-local"

local_app.py

@@ -0,0 +1,15 @@
+from local_plugin import SingingVoiceCloneLocal
+from tuneflow_devkit import Runner
+from pathlib import Path
+import uvicorn
+
+if __name__ == "__main__":
+    app = Runner(
+        plugin_class_list=[SingingVoiceCloneLocal],
+        bundle_file_path=str(Path(__file__).parent.joinpath('bundle_local.json').absolute())).start(
+        path_prefix='/plugins/singing-voice-clone-local',
+        config={
+            "corsConfig":
+                {"allowedOrigins": ["https://www.tuneflow.com", "https://tuneflow.com", "https://staging.tuneflow.com", "http://127.0.0.1:5173"],
+                 "addAllowCredentialsHeader": True}})
+    uvicorn.run(app)

local_debug.py

@@ -0,0 +1,7 @@
+from local_plugin import SingingVoiceCloneLocal
+from tuneflow_devkit import Debugger
+from pathlib import Path
+
+if __name__ == "__main__":
+    Debugger(plugin_class=SingingVoiceCloneLocal, bundle_file_path=str(
+        Path(__file__).parent.joinpath('bundle_local.json').absolute())).start()

local_plugin.py

@@ -5,9 +5,8 @@
 import traceback
 from inferencer import vc_fn_model, load_custom_model_func
 from utils import trim_audio
-from tuneflow_devkit import Runner
 from pathlib import Path
-import uvicorn
+
 
 class SingingVoiceCloneLocal(TuneflowPlugin):
     @staticmethod
@@ -155,9 +154,12 @@ def run(song: Song, params: dict[str, Any]):
 
         try:
             model, spk = load_custom_model_func(config_path=config_file, ckpt_path=model_file)
-            result = vc_fn_model(model, spk, trim_audio(
-            clip_audio_data_list[0]["audioData"]["data"], song, clip), vc_transform=pitchOffset, auto_f0=False, cluster_ratio=0, slice_db=-40,
-                           noise_scale=0.4, pad_seconds=0.5, cl_num=0, lg_num=0, lgr_num=0.75, F0_mean_pooling=f0MeanPooling, enhancer_adaptive_key=0, cr_threshold=f0Threshold)
+            result = vc_fn_model(
+                model, spk, trim_audio(clip_audio_data_list[0]["audioData"]["data"],
+                                       song, clip),
+                vc_transform=pitchOffset, auto_f0=False, cluster_ratio=0, slice_db=-40, noise_scale=0.4,
+                pad_seconds=0.5, cl_num=0, lg_num=0, lgr_num=0.75, F0_mean_pooling=f0MeanPooling,
+                enhancer_adaptive_key=0, cr_threshold=f0Threshold)
             if not result:
                 raise Exception("Failed to generate audio")
             status, generated_data = result
@@ -177,9 +179,4 @@ def run(song: Song, params: dict[str, Any]):
             }, clip_end_tick=clip.get_clip_end_tick(), insert_clip=True)
         except Exception as e:
             print(traceback.format_exc())
-            raise e
-
-if __name__ == "__main__":
-    app = Runner(plugin_class_list=[SingingVoiceCloneLocal], bundle_file_path=str(
-        Path(__file__).parent.joinpath('bundle_local.json').absolute())).start(path_prefix='/plugins/singing-voice-clone-local')
-    uvicorn.run(app)
+            raise e

requirements.txt

@@ -1,4 +1,4 @@
-tuneflow-devkit-py>=0.8.6
+tuneflow-devkit-py>=0.8.7
 # Flask
 # Flask_Cors
 # gradio

requirements.win.txt

@@ -1,4 +1,4 @@
-tuneflow-devkit-py>=0.8.6
+tuneflow-devkit-py>=0.8.7
 librosa==0.9.2
 fairseq==0.12.2
 Flask==2.1.2

utils.py

@@ -7,20 +7,17 @@
 import json
 import subprocess
 import warnings
-import random
 import functools
 
-import librosa
 import numpy as np
 from scipy.io.wavfile import read
 import torch
 from torch.nn import functional as F
-from modules.commons import sequence_mask
-from hubert import hubert_model
 from pathlib import Path
 from tuneflow_py import Song, Clip
 from pydub import AudioSegment
 from io import BytesIO
+from fairseq.checkpoint_utils import load_model_ensemble_and_task
 
 MATPLOTLIB_FLAG = False
 
@@ -214,8 +211,7 @@ def f0_to_coarse(f0):
 def get_hubert_model():
   vec_path = str(Path(__file__).parent.joinpath('hubert').joinpath('checkpoint_best_legacy_500.pt'))
   print("load model(s) from {}".format(vec_path))
-  from fairseq import checkpoint_utils
-  models, saved_cfg, task = checkpoint_utils.load_model_ensemble_and_task(
+  models, saved_cfg, task = load_model_ensemble_and_task(
     [vec_path],
     suffix="",
   )