1
00:00:00,000 --> 00:00:08,000
嘿,大家好,我莫尉。給大家介紹一下我搓了一個多月搓出來的一個巨牛,巨厲害的一個字幕編輯器。

2
00:00:08,000 --> 00:00:20,000
準確的說是一個視頻的字幕生成工作流。它可以幹什麼呢?它是這樣子的。首先你可以看到右邊這個非常厲害的多行的這個波形。

3
00:00:20,000 --> 00:00:28,000
點到哪裡就可以跳出來。這樣比起傳統的單行波形。單行波形我們只能看到這麼多,滾動起來很慢。

4
00:00:28,000 --> 00:00:29,760
好我们可以一下子看到这么多

5
00:00:29,760 --> 00:00:31,400
相当于涨了12345678

6
00:00:31,400 --> 00:00:32,140
10倍

7
00:00:32,140 --> 00:00:34,220
对10倍容量跳起来就快很多

8
00:00:34,220 --> 00:00:36,300
当然这个需要声明

9
00:00:36,300 --> 00:00:38,520
这个是从之前另一个up主分享的

10
00:00:38,520 --> 00:00:40,720
gap gone那里超过来的

11
00:00:40,720 --> 00:00:42,640
那边也是一个开源项目

12
00:00:42,640 --> 00:00:44,300
所以就直接借鉴了

13
00:00:44,300 --> 00:00:45,940
然后我这也是一个开源项目

14
00:00:45,940 --> 00:00:48,100
然后呢我们可以跳一下看看

15
00:00:48,100 --> 00:00:49,180
好的书街上回

16
00:00:49,180 --> 00:00:50,640
该叫什么呢

17
00:00:50,640 --> 00:00:51,340
牧羊人

18
00:00:51,340 --> 00:00:52,640
好这里看到一个

19
00:00:52,640 --> 00:00:56,120
好这个跳转播放都是基础功能

20
00:00:56,120 --> 00:00:57,220
没什么特别的对吧

21
00:00:57,220 --> 00:00:59,180
接下来介绍一些特别的

22
00:00:59,180 --> 00:01:01,380
也就是我这个工作流

23
00:01:01,380 --> 00:01:04,100
其实是使用了千问的ASR生成的

24
00:01:04,100 --> 00:01:06,700
然后一方面它的精度非常高

25
00:01:06,700 --> 00:01:08,800
这里面其实就是原生的精度

26
00:01:08,800 --> 00:01:11,540
你可以看到基本上甚至它代表点符号

27
00:01:11,540 --> 00:01:13,280
而且它没有什么错别字

28
00:01:13,280 --> 00:01:15,000
很少需要手工修的

29
00:01:15,000 --> 00:01:18,780
我实测下来可能95%以上都是准确的字词

30
00:01:18,780 --> 00:01:22,780
然后ASR有一个千问ASR有一个好处

31
00:01:22,780 --> 00:01:25,520
就是它可以生成字词级别的时间码

32
00:01:25,520 --> 00:01:27,300
意思就是你看这一句

33
00:01:27,300 --> 00:01:29,120
比如说我觉得它时间比较长

34
00:01:29,120 --> 00:01:30,720
我在这里拆分

35
00:01:30,720 --> 00:01:33,320
你有没有发现它不是像传统的拆分一样

36
00:01:33,320 --> 00:01:34,900
直接在这里中间割一刀

37
00:01:34,900 --> 00:01:37,940
然后前后的字可能还需要自己调

38
00:01:37,940 --> 00:01:38,880
而是直接

39
00:01:38,880 --> 00:01:45,600
我们再来演示一遍

40
00:01:45,600 --> 00:01:46,920
比如说这句

41
00:01:46,920 --> 00:01:48,360
这句我想在中间拆

42
00:01:48,360 --> 00:01:49,880
不对这句可能不太好拆

43
00:01:49,880 --> 00:01:51,300
因为它本身中间没有空行

44
00:01:51,300 --> 00:01:53,220
没有空格

45
00:01:53,220 --> 00:01:55,660
我还不知道是学校的教职人

46
00:01:55,660 --> 00:01:57,100
太准了

47
00:01:57,100 --> 00:01:59,640
就是它这种中间的小空隙

48
00:01:59,640 --> 00:02:01,120
它也可以直接拆掉

49
00:02:01,120 --> 00:02:02,700
然后如果我们这里觉得比较多

50
00:02:02,700 --> 00:02:04,260
直接往里拖一下

51
00:02:04,260 --> 00:02:07,000
这些都是直接待会儿导出字幕

52
00:02:07,000 --> 00:02:08,280
就可以保存下来的

53
00:02:08,280 --> 00:02:11,480
另一点就是在左边其实还有字幕列表

54
00:02:11,480 --> 00:02:13,120
比如说我不想看这边

55
00:02:13,120 --> 00:02:14,700
我想就按文字来看

56
00:02:14,700 --> 00:02:16,340
我们可以编剧视图

57
00:02:16,340 --> 00:02:18,600
然后把字幕拖过来

58
00:02:18,600 --> 00:02:21,700
这样子就有一个雷霆巨大的字幕区域了

59
00:02:21,700 --> 00:02:23,020
也可以这样拖上来

60
00:02:23,020 --> 00:02:24,020
这样子的话

61
00:02:25,220 --> 00:02:26,180
OK就这个试图吧

62
00:02:26,180 --> 00:02:26,900
完成布局

63
00:02:26,900 --> 00:02:27,580
这样子的话呢

64
00:02:27,580 --> 00:02:30,340
我们就可以直接按这个文本来跳

65
00:02:30,340 --> 00:02:33,460
因为有时候可能文本更容易定位

66
00:02:33,460 --> 00:02:34,300
比如说这里

67
00:02:34,300 --> 00:02:36,420
这边有一本很红的书

68
00:02:36,420 --> 00:02:37,580
很红的书对吧

69
00:02:37,580 --> 00:02:38,740
我在这里我想拆

70
00:02:39,820 --> 00:02:40,620
也是一样的

71
00:02:40,620 --> 00:02:41,780
它会直接拆掉

72
00:02:41,780 --> 00:02:43,660
有一本很红的书

73
00:02:43,660 --> 00:02:44,500
看来是我的

74
00:02:44,500 --> 00:02:46,180
而且呢有一个很巧妙的设计

75
00:02:46,180 --> 00:02:47,900
就是在右边的波形

76
00:02:47,900 --> 00:02:49,860
我们是按音频去拆分的

77
00:02:49,860 --> 00:02:51,620
但是在左边的文字呢

78
00:02:51,620 --> 00:02:53,400
我们就算文字位置去拆分的

79
00:02:53,400 --> 00:02:54,800
可能也就跟索菲关系比较

80
00:02:54,800 --> 00:02:56,500
比如说这里都没有交流的

81
00:02:56,500 --> 00:02:57,680
我在这里右键拆

82
00:02:57,680 --> 00:03:00,380
它就是从都没有交流的

83
00:03:00,380 --> 00:03:01,940
可能也就跟索菲关系比较

84
00:03:01,940 --> 00:03:02,640
看出来了吧

85
00:03:02,640 --> 00:03:04,380
这是我觉得特别厉害

86
00:03:04,380 --> 00:03:05,300
特别方便的一点

87
00:03:05,300 --> 00:03:07,880
就是首先它大部分的时长

88
00:03:07,880 --> 00:03:08,480
其实你看

89
00:03:08,480 --> 00:03:11,100
我这里时长设的是一行16个字

90
00:03:11,100 --> 00:03:13,140
大部分其实都是在16个字之内的

91
00:03:13,140 --> 00:03:13,980
只有少部分

92
00:03:13,980 --> 00:03:15,180
哪怕超一点

93
00:03:15,180 --> 00:03:18,020
这里看到一个很科技感的

94
00:03:18,020 --> 00:03:19,140
我们可以在这里

95
00:03:19,140 --> 00:03:21,240
比如说设成20个字

96
00:03:21,240 --> 00:03:22,320
20个字之后

97
00:03:22,320 --> 00:03:24,340
超出20个字的就不多了

98
00:03:24,340 --> 00:03:25,620
我们可以紧看超长

99
00:03:25,620 --> 00:03:26,840
就只有这么几句

100
00:03:26,840 --> 00:03:31,980
休息的我在这里拆开

101
00:03:31,980 --> 00:03:33,200
先暂停一小时

102
00:03:33,200 --> 00:03:35,100
你看它时常没有超

103
00:03:35,100 --> 00:03:36,320
它就不显示了

104
00:03:36,320 --> 00:03:37,680
定时提醒休息的

105
00:03:37,680 --> 00:03:38,540
先暂停了一小时

106
00:03:38,540 --> 00:03:39,060
然后呢

107
00:03:39,060 --> 00:03:39,980
我也可以把它

108
00:03:39,980 --> 00:03:41,220
再点击

109
00:03:41,220 --> 00:03:42,420
点击跳转过来之后

110
00:03:42,420 --> 00:03:43,700
如果我想把这三句合并

111
00:03:43,700 --> 00:03:44,720
那么全选

112
00:03:44,720 --> 00:03:46,680
然后右键合并三条字幕

113
00:03:46,680 --> 00:03:47,600
需要把我一个

114
00:03:47,600 --> 00:03:49,000
定时提醒休息的

115
00:03:49,000 --> 00:03:49,940
先暂停了一小时

116
00:03:49,940 --> 00:03:50,240
OK

117
00:03:50,240 --> 00:03:50,920
好

118
00:03:50,920 --> 00:03:53,460
然后如果想在这里拆

119
00:03:53,460 --> 00:03:54,200
你看这句

120
00:03:54,200 --> 00:03:56,260
它的音频波形不太明显

121
00:03:56,260 --> 00:03:57,320
我就直接在这里

122
00:03:57,320 --> 00:03:58,460
找到这个空的位置

123
00:03:58,460 --> 00:03:59,080
开分

124
00:03:59,080 --> 00:04:01,720
如果我想在定时前面

125
00:04:01,720 --> 00:04:02,020
开分

126
00:04:02,020 --> 00:04:04,280
它非常符合直觉

127
00:04:04,280 --> 00:04:05,420
懂我意思吧

128
00:04:05,420 --> 00:04:07,380
然后还有什么功能

129
00:04:07,380 --> 00:04:07,860
我看一下

130
00:04:07,860 --> 00:04:09,380
还有一个就是

131
00:04:09,380 --> 00:04:11,700
字幕的一个基础功能

132
00:04:11,700 --> 00:04:13,020
比如说这个游戏叫做

133
00:04:13,020 --> 00:04:13,700
要理问答

134
00:04:13,700 --> 00:04:15,780
然后有这个

135
00:04:15,780 --> 00:04:17,180
它不太识别的出来

136
00:04:17,180 --> 00:04:18,760
所以我要是搜要理

137
00:04:18,760 --> 00:04:20,360
只有一个地方

138
00:04:20,360 --> 00:04:23,420
对 药理 我可以给它替换成药理

139
00:04:23,420 --> 00:04:25,140
这样子就可以直接预览到

140
00:04:25,140 --> 00:04:26,660
它会把这一句替换掉

141
00:04:26,660 --> 00:04:28,140
替换全部就可以了

142
00:04:28,140 --> 00:04:30,900
然后像这种修改过的或者替换过的

143
00:04:30,900 --> 00:04:32,560
都会有一个黄色的标记

144
00:04:32,560 --> 00:04:33,940
说明你改动过

145
00:04:33,940 --> 00:04:34,860
有什么来着

146
00:04:34,860 --> 00:04:36,280
还有一个重量级的东西

147
00:04:36,280 --> 00:04:36,820
待会介绍

148
00:04:36,820 --> 00:04:38,620
先继续介绍皮量替换

149
00:04:38,620 --> 00:04:41,540
比如说我把所有的我都替换成我

150
00:04:41,540 --> 00:04:42,900
这样子就

151
00:04:42,900 --> 00:04:44,840
所有的我这里都会变成

152
00:04:44,840 --> 00:04:46,480
MG的我了

153
00:04:46,480 --> 00:04:47,760
虽然效果有点雷霆

154
00:04:47,760 --> 00:04:49,020
没有关系可以撤销

155
00:04:49,020 --> 00:04:50,260
撤销掉就可以了

156
00:04:50,260 --> 00:04:53,060
然后还有一个特别厉害的功能

157
00:04:53,060 --> 00:04:53,700
移除静音

158
00:04:53,700 --> 00:04:56,280
这个也是从GapGone那边参考过来的

159
00:04:56,280 --> 00:04:58,860
但是我加了一些参数

160
00:04:58,860 --> 00:05:00,540
高级参数我们先不管它

161
00:05:00,540 --> 00:05:02,820
我们直接点击扫描并移除

162
00:05:02,820 --> 00:05:04,300
你可以看到这些橙色的部分

163
00:05:04,300 --> 00:05:06,200
都是它识别到的空隙

164
00:05:06,200 --> 00:05:09,020
然后会被自动移除掉

165
00:05:09,020 --> 00:05:11,720
你看全长40多分钟吧

166
00:05:11,720 --> 00:05:14,640
其实里面有整整25分钟的空隙

167
00:05:14,640 --> 00:05:15,620
占了一半以上

168
00:05:15,620 --> 00:05:16,760
我们在这个时候播放呢

169
00:05:16,760 --> 00:05:17,640
它就会自动跳过

170
00:05:17,640 --> 00:05:27,400
对吧很厉害吧

171
00:05:27,400 --> 00:05:28,920
然后如果说比如说

172
00:05:28,920 --> 00:05:29,700
这一段

173
00:05:29,700 --> 00:05:32,300
这一段我是要保留的

174
00:05:32,300 --> 00:05:33,260
因为这是个笑声

175
00:05:33,260 --> 00:05:35,100
所以我可以按住alt点它

176
00:05:35,100 --> 00:05:38,480
这样就可以保留了

177
00:05:38,480 --> 00:05:40,000
然后如果要重新静音就这样

178
00:05:40,000 --> 00:05:41,900
而且它还可以直接

179
00:05:41,900 --> 00:05:44,220
中间拖动时添加到这个静音片段

180
00:05:44,220 --> 00:05:45,620
比如说这里我都不需要了

181
00:05:45,620 --> 00:05:47,360
就直接这样子

182
00:05:47,360 --> 00:05:48,680
他就从前面播

183
00:05:48,680 --> 00:05:49,440
他摔下来了

184
00:05:49,440 --> 00:05:51,180
还真是直接跳过

185
00:05:51,180 --> 00:05:52,560
当然为了预览

186
00:05:52,560 --> 00:05:54,460
如果你点到中间去播的话

187
00:05:54,460 --> 00:05:56,600
他就是会照播不误

188
00:05:56,600 --> 00:05:58,080
是我老师

189
00:05:58,080 --> 00:05:59,060
是我放了个小

190
00:05:59,060 --> 00:06:00,100
然后如果说

191
00:06:00,100 --> 00:06:01,440
这个我又想恢复呢

192
00:06:01,440 --> 00:06:02,060
我按住alt

193
00:06:02,060 --> 00:06:03,000
然后中间拖动

194
00:06:03,000 --> 00:06:04,720
就可以给他恢复回来了

195
00:06:04,720 --> 00:06:07,220
还真是互相

196
00:06:07,220 --> 00:06:07,960
大概这样

197
00:06:07,960 --> 00:06:08,620
但是呢

198
00:06:08,620 --> 00:06:09,480
就是你说

199
00:06:09,480 --> 00:06:10,740
这个编辑起来很厉害

200
00:06:10,740 --> 00:06:11,740
那我们怎么用呢

201
00:06:11,740 --> 00:06:13,880
在这里导出字幕

202
00:06:13,880 --> 00:06:18,380
就是把你编辑完的这个字幕给导出成传统的SRT

203
00:06:18,380 --> 00:06:20,560
然后你可以导出空隙版本呢

204
00:06:20,560 --> 00:06:24,480
就是这个按静音轴剪过的这个字幕

205
00:06:24,480 --> 00:06:27,800
然后这里OTIO其实是OpenTimeline的那个工程

206
00:06:27,800 --> 00:06:29,260
我是用来给达芬奇用的

207
00:06:29,260 --> 00:06:31,440
目前因为我自己用的是达芬奇

208
00:06:31,440 --> 00:06:34,420
所以目前还没有其他软件的工程支持

209
00:06:34,420 --> 00:06:39,240
如果反响好的话之后再看吧

210
00:06:39,240 --> 00:06:41,260
比如说剪音什么的给弄一下

211
00:06:41,260 --> 00:06:42,580
然后这两个呢

212
00:06:42,580 --> 00:06:45,480
其实就是如果你自己AI会用

213
00:06:45,480 --> 00:06:46,880
或者有一定编程能力的话

214
00:06:46,880 --> 00:06:49,720
这两个也是相当于都是告诉你

215
00:06:49,720 --> 00:06:53,200
这个视频里面哪些时间区域保留了的

216
00:06:53,200 --> 00:06:55,060
你把这任意一个

217
00:06:55,060 --> 00:06:57,100
第一个是给FFMPG用的

218
00:06:57,100 --> 00:06:58,800
第二个就是纯数据JSON

219
00:06:58,800 --> 00:07:00,140
你可以把它保存下来

220
00:07:00,140 --> 00:07:02,020
就自己让AI写个脚本

221
00:07:02,020 --> 00:07:03,880
去用来切分视频

222
00:07:03,880 --> 00:07:04,420
也没问题

223
00:07:04,420 --> 00:07:07,460
因为我这里只是提供一个交互式的编辑界面

224
00:07:07,460 --> 00:07:11,400
然后其实还有一点比较有趣的功能

225
00:07:11,400 --> 00:07:13,160
比如说我们来举个例子

226
00:07:13,160 --> 00:07:16,180
那这句我可以给他右键红色

227
00:07:16,180 --> 00:07:19,800
然后或者说所有人都

228
00:07:19,800 --> 00:07:22,420
这几句我可以给他批量选择用红色

229
00:07:22,420 --> 00:07:23,080
那我就白

230
00:07:23,080 --> 00:07:26,120
那人就要问了这个字幕的颜色有什么用呢

231
00:07:26,120 --> 00:07:28,260
那一方面在这里编辑的时候

232
00:07:28,260 --> 00:07:30,920
有颜色可以让自己当个标记吧

233
00:07:30,920 --> 00:07:31,700
另一方面呢

234
00:07:31,700 --> 00:07:34,840
这个颜色是可以导出成达芬奇的Jason的

235
00:07:34,840 --> 00:07:36,240
然后我有一个小本

236
00:07:36,240 --> 00:07:37,420
可以在达芬奇中

237
00:07:37,420 --> 00:07:38,720
也给这些字幕

238
00:07:38,720 --> 00:07:40,280
添加上片段颜色

239
00:07:40,280 --> 00:07:41,220
这样的好处呢

240
00:07:41,220 --> 00:07:42,980
就是我可以快速的知道

241
00:07:42,980 --> 00:07:44,520
在哪些地方

242
00:07:44,520 --> 00:07:45,940
比如说我把红色

243
00:07:45,940 --> 00:07:47,220
认为是重点内容

244
00:07:47,220 --> 00:07:49,060
那那部分可能就要加一些

245
00:07:49,060 --> 00:07:50,940
画面的放大效果之类的

246
00:07:50,940 --> 00:07:52,520
然后紫色可能是

247
00:07:52,520 --> 00:07:53,620
比较有乐子的地方

248
00:07:53,620 --> 00:07:54,820
那那些地方也可以加

249
00:07:54,820 --> 00:07:56,040
对应的音乐啥的

250
00:07:56,040 --> 00:07:56,600
总之就是

251
00:07:56,600 --> 00:07:58,580
因为我自己在编辑的时候

252
00:07:58,580 --> 00:07:59,660
注意一下体力空

253
00:08:00,000 --> 00:08:07,000
因為我自己在編輯的時候,這一次編輯下來,相當於First Bus,它不光是一個字幕的編輯,它還是我對整體這個視頻怎麼剪。

254
00:08:07,000 --> 00:08:08,680
比如说如果是这一段

255
00:08:08,680 --> 00:08:10,520
我都觉得不想要

256
00:08:10,520 --> 00:08:11,940
可以直接这个呢

257
00:08:11,940 --> 00:08:13,180
加速跳过的

258
00:08:13,180 --> 00:08:15,200
那么我给它全选中之后

259
00:08:15,200 --> 00:08:16,880
比如说标个绿色吧

260
00:08:16,880 --> 00:08:17,740
绿色就当跳过

261
00:08:17,740 --> 00:08:19,480
OK看起来就这样子

262
00:08:19,480 --> 00:08:21,180
到时候我在达芬奇工程里面

263
00:08:21,180 --> 00:08:21,800
就可以知道

264
00:08:21,800 --> 00:08:24,440
这一部分是需要一起跳过的

265
00:08:24,440 --> 00:08:26,600
除了颜色还有一个

266
00:08:26,600 --> 00:08:28,940
就是它可以分配表情包

267
00:08:28,940 --> 00:08:32,120
这个表情包

268
00:08:32,120 --> 00:08:33,280
我觉得

269
00:08:33,280 --> 00:08:35,620
这个呢你行不行

270
00:08:35,620 --> 00:08:36,200
应该

271
00:08:36,200 --> 00:08:36,980
我看你啊

272
00:08:36,980 --> 00:08:39,020
其他人能用上的可能不多

273
00:08:39,020 --> 00:08:40,380
所以目前这个功能

274
00:08:40,380 --> 00:08:41,980
我还没有放到公开仓库里面

275
00:08:41,980 --> 00:08:44,980
就是也是可以导出成一个OTIO工程

276
00:08:44,980 --> 00:08:46,980
然后快速的导入达芬奇

277
00:08:46,980 --> 00:08:48,780
就可以在这句里面

278
00:08:48,780 --> 00:08:50,600
配上对应的表情包

279
00:08:50,600 --> 00:08:54,900
这个看有没有别人有需求吧

280
00:08:54,900 --> 00:08:57,460
因为我觉得喜欢往字幕上配表情包

281
00:08:57,460 --> 00:08:58,480
一个是工作量比较大

282
00:08:58,480 --> 00:09:01,140
另一个是我这种表情包狂魔

283
00:09:01,140 --> 00:09:02,500
会对这种需求特别

284
00:09:02,500 --> 00:09:05,020
那绿下面有两位的话

285
00:09:05,020 --> 00:09:08,860
目前我是把这个先发了一个Github的仓库

286
00:09:08,860 --> 00:09:11,300
相当于是完全复刻我现有的工作流

287
00:09:11,300 --> 00:09:13,940
所以它会有一点技术要求

288
00:09:13,940 --> 00:09:15,580
它需要你电脑上有python

289
00:09:15,580 --> 00:09:19,620
然后需要自己准备一个签问ASR的API

290
00:09:19,620 --> 00:09:22,720
然后就可以生成这样一个网页

291
00:09:22,720 --> 00:09:24,440
它其实就是个HTML网页

292
00:09:24,440 --> 00:09:27,300
你本地浏览器打开就可以看

293
00:09:27,300 --> 00:09:30,060
然后我之后有打算现在这个工程

294
00:09:30,060 --> 00:09:32,060
现在这个项目叫做Maw

295
00:09:32,060 --> 00:09:34,820
Mois ASR Workflow

296
00:09:34,820 --> 00:09:37,180
之后我又打算做成一个

297
00:09:37,180 --> 00:09:38,920
MOSE

298
00:09:38,920 --> 00:09:41,240
Mois Open Subtitle Editor

299
00:09:41,240 --> 00:09:41,960
墨色

300
00:09:41,960 --> 00:09:43,480
墨色字幕编辑器

301
00:09:43,480 --> 00:09:44,980
到那时候应该

302
00:09:44,980 --> 00:09:47,480
我会想办法把它封装成

303
00:09:47,480 --> 00:09:49,980
更简单易用的一个形式

304
00:09:49,980 --> 00:09:50,720
当然也有可能

305
00:09:50,720 --> 00:09:52,740
如果这个视频反响频频的话

306
00:09:52,740 --> 00:09:54,460
那我也就懒得折腾了

307
00:09:54,460 --> 00:09:56,540
因为我自己现在用这个东西

308
00:09:56,540 --> 00:09:58,040
已经是用得非常爽了

309
00:09:58,040 --> 00:09:59,500
包括你们现在看到的这个视频

310
00:09:59,500 --> 00:10:00,460
我也是用它

311
00:10:00,460 --> 00:10:03,480
我也将会用它来做剪辑工作

312
00:10:03,480 --> 00:10:05,020
剪辑的前期工作吧

313
00:10:05,020 --> 00:10:06,480
因为这里处理完之后

314
00:10:06,480 --> 00:10:07,240
导入达芬奇

315
00:10:07,240 --> 00:10:08,660
就字幕也有了

316
00:10:08,660 --> 00:10:10,820
然后空隙去除也有

317
00:10:10,820 --> 00:10:11,580
就很方便

318
00:10:11,580 --> 00:10:12,120
好

319
00:10:12,120 --> 00:10:13,740
这次是一个相当于

320
00:10:13,740 --> 00:10:14,880
我忍不住

321
00:10:14,880 --> 00:10:16,980
忍不住先介绍一下

322
00:10:16,980 --> 00:10:17,980
所以没有太精简

323
00:10:17,980 --> 00:10:21,180
先纯续续刀的一个视频

324
00:10:21,180 --> 00:10:23,260
相当于先预告一下吧

325
00:10:23,260 --> 00:10:24,480
等之后有正式发布了

326
00:10:24,480 --> 00:10:27,020
会再做正式的介绍事

327
00:10:27,020 --> 00:10:27,560
也许吧

328
00:10:27,560 --> 00:10:28,260
如果没有后续

329
00:10:28,260 --> 00:10:29,100
那就没有后续了

330
00:10:29,100 --> 00:10:31,320
然后现在原代码都是开放的

331
00:10:31,320 --> 00:10:32,700
所以你也可以拿过去

332
00:10:32,700 --> 00:10:35,160
然后按自己的习惯或者偏好

333
00:10:35,160 --> 00:10:37,420
用AI改或者自己改都行

334
00:10:37,420 --> 00:10:38,500
用的愉快吧

335
00:10:38,500 --> 00:10:41,860
因为我是希望能给其他游戏UP主

336
00:10:41,860 --> 00:10:43,140
也带来一些帮助的

337
00:10:43,140 --> 00:10:43,860
大家都知道

338
00:10:43,860 --> 00:10:47,240
这个时代还在做这种传统游戏实况的

339
00:10:47,240 --> 00:10:48,460
其实不多了

340
00:10:48,460 --> 00:10:52,020
所以希望给大家能节约一些后期的时间吧

341
00:10:52,020 --> 00:10:56,080
我是知道整这个字幕有多麻烦的

342
00:10:56,080 --> 00:10:58,940
所以在节约我自己时间的时候

343
00:10:58,940 --> 00:11:00,700
也希望给大家节约一点时间

344
00:11:00,700 --> 00:11:01,280
对了

345
00:11:01,280 --> 00:11:02,740
我突然意识到我漏了一个东西

346
00:11:02,740 --> 00:11:04,500
这个东西我还没放进库里

347
00:11:04,500 --> 00:11:06,320
但是它还蛮厉害的

348
00:11:06,320 --> 00:11:07,620
就是你看这种

349
00:11:07,620 --> 00:11:09,240
这里有一些灰色的字幕

350
00:11:09,240 --> 00:11:09,920
代表什么呢

351
00:11:09,920 --> 00:11:11,200
我们来播一下看看

352
00:11:11,200 --> 00:11:12,480
这是一遍

353
00:11:12,480 --> 00:11:14,000
好菜

354
00:11:14,000 --> 00:11:14,940
这是什么情况

355
00:11:14,940 --> 00:11:17,040
我的心快要从胸口里蹦出来了

356
00:11:17,040 --> 00:11:19,600
我吸收了它的铁

357
00:11:19,600 --> 00:11:21,140
文字机

358
00:11:21,140 --> 00:11:22,980
你会看到灰色的这个

359
00:11:22,980 --> 00:11:25,260
其实是这个项目的一个附属工作流

360
00:11:25,260 --> 00:11:27,320
就是我先生成了所有的字幕

361
00:11:27,320 --> 00:11:30,400
然后用字幕的每一句的这个时间

362
00:11:30,400 --> 00:11:33,580
在视频里面截取画面的这个部分

363
00:11:33,580 --> 00:11:34,920
进行OCR识别

364
00:11:34,920 --> 00:11:38,640
如果识别到画面上的台词内容

365
00:11:38,640 --> 00:11:42,780
跟我字幕内容是相似度大于百分之多少

366
00:11:42,780 --> 00:11:45,100
就认为是对着画面配音

367
00:11:45,100 --> 00:11:48,200
那这些就直接灰掉了

368
00:11:48,200 --> 00:11:49,720
当然我也可以重新给它启用

369
00:11:49,720 --> 00:11:51,080
这本书上有一张字条

370
00:11:51,080 --> 00:11:52,720
这本书上有一张字

371
00:11:52,720 --> 00:11:54,460
就重新启用的话就有字幕

372
00:11:54,460 --> 00:11:56,000
但是禁用掉的话

373
00:11:56,000 --> 00:11:57,140
就是导出字幕的时候

374
00:11:57,140 --> 00:11:58,320
不会导出这句

375
00:11:58,320 --> 00:11:59,060
这样可以避免

376
00:11:59,060 --> 00:12:01,220
相当于字幕的文本

377
00:12:01,220 --> 00:12:03,020
跟游戏本身的文本

378
00:12:03,020 --> 00:12:04,280
叠在一起这种情况

379
00:12:04,280 --> 00:12:06,300
这个我觉得也是

380
00:12:06,300 --> 00:12:08,480
像我这样的老辈子

381
00:12:08,480 --> 00:12:10,720
配音游戏实况UP主

382
00:12:10,720 --> 00:12:12,440
会用到的一个功能吧

383
00:12:12,440 --> 00:12:14,660
但是因为它又涉及到多一个

384
00:12:14,660 --> 00:12:16,260
这个是用了Paddle

385
00:12:16,260 --> 00:12:18,040
百度的Paddle OCR

386
00:12:18,040 --> 00:12:19,380
流程更复杂一点

387
00:12:19,380 --> 00:12:20,980
所以暂时没有放进

388
00:12:20,980 --> 00:12:21,980
这个公开版本中

389
00:12:21,980 --> 00:12:23,020
后续再看吧

390
00:12:23,020 --> 00:12:24,620
OK就介绍到这里

391
00:12:24,620 --> 00:12:25,860
感谢再见

392
00:12:26,000 --> 00:12:26,260
拜拜

393
00:12:26,260 --> 00:12:27,320
如果有兴趣的话

394
00:12:27,320 --> 00:12:28,800
可以关注一下

395
00:12:28,800 --> 00:12:30,820
因为这个项目

396
00:12:30,820 --> 00:12:32,840
后续会进行到

397
00:12:32,840 --> 00:12:34,260
什么样的状态

398
00:12:34,260 --> 00:12:34,920
就完全看

399
00:12:34,920 --> 00:12:36,400
有多少人需要了

400
00:12:36,400 --> 00:12:38,120
如果发现是

401
00:12:38,120 --> 00:12:39,260
纯我自己自嗨

402
00:12:39,260 --> 00:12:40,660
觉得这个有用的人

403
00:12:40,660 --> 00:12:41,180
不多的话

404
00:12:41,180 --> 00:12:41,920
那可能也就

405
00:12:41,920 --> 00:12:42,680
自己继续用

406
00:12:42,680 --> 00:12:43,680
OK就到这儿

407
00:12:43,680 --> 00:12:44,080
下次见

408
00:12:44,080 --> 00:12:44,420
拜拜
