bump version

README.md

@@ -23,12 +23,10 @@ you meet with your ```log.log``` file on github issues.
 2. Download latest version of Freyja from [Releases](https://github.com/freyjaSubOCR/freyja-sub-ocr-electron/releases)
    page and extract it.
 
-3. Download models from <https://github.com/freyjaSubOCR/freyja-sub-ocr-model-zoo/releases>. Choose GPU models if you
-   have a Nvidia GPU, otherwise choose CPU models. Download all txt and torchscript files, and place these files into
-   ```<app_path>/models/``` folder.
+3. Download models from <https://github.com/freyjaSubOCR/freyja-sub-ocr-model-zoo/releases>. Download all txt and
+   torchscript files, and place these files into ```<app_path>/models/``` folder.
 
-4. Run ```freyja.exe```. Enable ```Enable CUDA``` option if you have a Nvidia GPU and downloaded GPU models,
-   otherwise disable the option.
+4. Run ```freyja.exe```. Enable ```Enable CUDA``` option if you have a Nvidia GPU, otherwise disable the option.
 
 ## Known issues
 

README_zh-hans.md

@@ -20,10 +20,10 @@ Freyja 需要 8GB 的内存。强烈推荐使用带 Nvidia 显卡的电脑，否
 
 2. 从[Release页](https://github.com/freyjaSubOCR/freyja-sub-ocr-electron/releases)下载最新版本的Freyja并将其解压缩。
 
-3. 从<https://github.com/freyjaSubOCR/freyja-sub-ocr-model-zoo/releases>中下载模型。如果你有 Nvidia 显卡，就下载GPU模型；
-   否则就下载CPU模型。下载对应模型的所有txt和torchscript文件，并将这些文件放入```<app_path>/models/```文件夹。
+3. 从<https://github.com/freyjaSubOCR/freyja-sub-ocr-model-zoo/releases>中下载模型。下载对应模型的所有txt和torchscript文
+   件，并将这些文件放入```<app_path>/models/```文件夹。
 
-4. 运行```freyja.exe```。如果有 Nvidia 显卡并下载了 GPU 模型，请启用```Enable CUDA```选项，否则禁用该选项。
+4. 运行```freyja.exe```。如果有 Nvidia 显卡，请启用```Enable CUDA```选项，否则禁用该选项。
 
 ## 已知的问题
 

changelog.md

@@ -0,0 +1,18 @@
+# Change Log
+
+## V0.4.0
+
+This version is updated to use OCRV3 models. OCRV3 models are faster than the old OCR models with little accuracy loss.
+The new models use less GPU memory, so the default batch size has been updated from 8 to 32. On a Surface book 2 laptop
+(i7-8650U, GTX 1060 Max-Q), it only takes 10 minutes to process a 24 minutes video.
+
+This version does not need an object detection model, and it unifies CPU and GPU models.
+
+To archive a higher accuracy, this version requires you to select a tighter subtitle boundary.
+
+这个版本使用了更新后的OCRV3模型。新的OCRV3模型相比旧的OCR模型来说运行速度更快，也更加准确。新模型使用的GPU内存较少，所以
+默认的批次大小从8个变更为32个。在Surface book 2笔记本上（i7-8650U，GTX 1060 Max-Q），处理一段24分钟的视频只需要10分钟。
+
+这个版本不需要以前的对象检测模型，也统一了CPU和GPU模型。
+
+新的模型需要你框选更准确的字幕边界，否则准确度会很差。

package.json

@@ -1,6 +1,6 @@
 {
     "name": "freyja",
-    "version": "0.3.0",
+    "version": "0.4.0",
     "author": "arition, ReventonC",
     "description": "freyja subtitle OCR extractor",
     "license": "GPL-3.0-or-later",