1
00:00:00,000 --> 00:00:25,200
大家好,這裡是Hex電腦核糖。本期要跟大家分享的是一款開源的同聲傳異的神器。那這款工具在我自己使用體驗後呢,我發現它的效果真的很驚艷。它不只是跨平台,支持Windows、Ninogus還有MacOS,並且整個過程是完全免費的。更重要的是它的效果是我覺得是可以跟很多商用必源的去幫忙手腕的。OK,但我們廢話不多說,先給大家直接展示效果。

2
00:00:25,200 --> 00:00:30,020
人工智能是一项对社会影响非常深远的技术

3
00:00:30,020 --> 00:00:32,220
它将重塑社会的各行各业

4
00:00:32,220 --> 00:00:48,300
我们从现在开始就要密切地关注这项技术

5
00:00:48,300 --> 00:00:51,120
并且保持与时俱进

6
00:00:51,120 --> 00:01:02,780
所以光注Hex电脑和谈

7
00:01:02,780 --> 00:01:04,340
并且持续地学习

8
00:01:04,340 --> 00:01:06,480
让文艺一起在AI时代保持进步

9
00:01:16,560 --> 00:01:20,560
人工智能是一项对社会发展影响深远的技术

10
00:01:20,560 --> 00:01:23,380
它将重塑我们社会够行够业

11
00:01:23,380 --> 00:01:37,780
人工智能

12
00:01:37,780 --> 00:01:39,540
它是一项对社会发展

13
00:01:39,540 --> 00:01:40,860
影响深远的技术

14
00:01:40,860 --> 00:01:43,380
它将重塑我们社会够行够业

15
00:01:43,380 --> 00:01:59,660
OK 大家可以看到刚才的演示

16
00:01:59,660 --> 00:02:01,320
是不是它的效果非常好

17
00:02:01,320 --> 00:02:04,720
那本期呢 我就先带大家手把手了

18
00:02:04,720 --> 00:02:06,180
来完成这个项目的一个部署

19
00:02:06,180 --> 00:02:08,940
OK 那首先呢它的系统是有些要求的

20
00:02:08,940 --> 00:02:10,540
就是你是 W10 W11

21
00:02:10,540 --> 00:02:11,460
这个都没关系

22
00:02:11,460 --> 00:02:13,980
然后你的电脑如果没有GPU的话

23
00:02:13,980 --> 00:02:15,300
也是OK的

24
00:02:15,300 --> 00:02:16,260
在要素比较低

25
00:02:16,260 --> 00:02:18,260
但是如果你有GPU

26
00:02:18,260 --> 00:02:19,700
有音伟达的显感那更好了

27
00:02:19,700 --> 00:02:22,120
建议显存是大于8G以上

28
00:02:22,120 --> 00:02:25,780
然后你的内存最低要4G

29
00:02:25,780 --> 00:02:26,960
这是最低的

30
00:02:26,960 --> 00:02:28,740
然后你的磁帮空间的话

31
00:02:28,740 --> 00:02:30,120
是200兆以上

32
00:02:30,120 --> 00:02:31,720
这个都是很低的条件

33
00:02:31,720 --> 00:02:33,460
然后你的麦克风和延伸器

34
00:02:33,460 --> 00:02:34,240
要正常工作

35
00:02:34,240 --> 00:02:35,980
第一步我们去下载这个终端

36
00:02:35,980 --> 00:02:37,140
我们来到这个网页

37
00:02:37,140 --> 00:02:38,520
这是它的光团网页

38
00:02:38,520 --> 00:02:39,880
我们点击这个下载

39
00:02:39,880 --> 00:02:41,540
然后点击下载之后

40
00:02:41,540 --> 00:02:43,640
这个中端就会来下载到

41
00:02:43,640 --> 00:02:45,500
有这样一个exe的文件

42
00:02:45,500 --> 00:02:46,380
我们双击它

43
00:02:46,380 --> 00:02:47,420
然后点击下一步

44
00:02:47,420 --> 00:02:48,080
下一步安装

45
00:02:48,080 --> 00:02:50,020
就可以安装到本地了

46
00:02:50,020 --> 00:02:50,260
OK

47
00:02:50,260 --> 00:02:51,600
然后顺便提一下

48
00:02:51,600 --> 00:02:53,140
因为它们两种形式

49
00:02:53,140 --> 00:02:54,240
我们今天讲的是

50
00:02:54,240 --> 00:02:55,400
这种中端的形式

51
00:02:55,400 --> 00:02:56,940
还有一种是它浏览器的

52
00:02:56,940 --> 00:02:58,360
一个拓展插件的形式

53
00:02:58,360 --> 00:02:59,540
这个后面有机会

54
00:02:59,540 --> 00:03:00,300
我再给大家讲

55
00:03:00,300 --> 00:03:01,100
OK

56
00:03:01,100 --> 00:03:03,020
你中端穿好之后

57
00:03:03,020 --> 00:03:03,880
你桌面就会有

58
00:03:03,880 --> 00:03:04,960
这样一个logo图标

59
00:03:04,960 --> 00:03:06,480
我们现在双击打开它

60
00:03:06,480 --> 00:03:07,920
它就会来到这样一页面

61
00:03:07,920 --> 00:03:08,620
我们可以看到

62
00:03:08,620 --> 00:03:09,620
右上角有登录

63
00:03:09,620 --> 00:03:11,380
有字幕有设置等按钮

64
00:03:11,380 --> 00:03:13,600
我们这里的点击设置

65
00:03:13,600 --> 00:03:15,440
我们这里可以选的

66
00:03:15,440 --> 00:03:16,680
我说什么语言

67
00:03:16,680 --> 00:03:17,560
我这里选中文

68
00:03:17,560 --> 00:03:18,940
然后对方听到什么语言

69
00:03:18,940 --> 00:03:19,700
你可以去选的

70
00:03:19,700 --> 00:03:21,020
就是这里选英语

71
00:03:21,020 --> 00:03:22,820
你一开始进来

72
00:03:22,820 --> 00:03:24,100
这里都是没有什么东西的

73
00:03:24,100 --> 00:03:24,840
你要自己去选

74
00:03:24,840 --> 00:03:26,060
然后这些选的话

75
00:03:26,060 --> 00:03:27,660
也不是你可以任意去选的

76
00:03:27,660 --> 00:03:29,280
要根据你自己去配置的模型

77
00:03:29,280 --> 00:03:30,900
模型具体怎么配置

78
00:03:30,900 --> 00:03:31,820
我给大家说一下

79
00:03:31,820 --> 00:03:33,460
比如大家可以看到

80
00:03:33,460 --> 00:03:36,080
它这里总共是包括了三个模块

81
00:03:36,080 --> 00:03:38,340
这里比如说有语音识别

82
00:03:38,340 --> 00:03:40,780
有语音的文本的翻译

83
00:03:40,780 --> 00:03:41,740
还有语音的合成

84
00:03:41,740 --> 00:03:44,380
这几个模型是我自己下载的

85
00:03:44,380 --> 00:03:46,040
我为什么选择下载自己的模型

86
00:03:46,040 --> 00:03:48,180
我给大家说一下原因

87
00:03:48,180 --> 00:03:50,360
因为我自己做了很多的一些测试

88
00:03:50,360 --> 00:03:52,160
这第二步的话

89
00:03:52,160 --> 00:03:53,240
我们本期先不讲

90
00:03:53,240 --> 00:03:55,420
因为这个东西是跟插件相关的

91
00:03:55,420 --> 00:03:57,800
跟我们这一次内容不是很相关

92
00:03:57,800 --> 00:03:58,760
所以我们这边先过

93
00:03:58,760 --> 00:04:00,860
然后我们先看到第三步

94
00:04:00,860 --> 00:04:02,140
就是模型下载这里

95
00:04:02,140 --> 00:04:03,360
模型下载这里

96
00:04:03,360 --> 00:04:04,900
首先它分的三个模型

97
00:04:04,900 --> 00:04:06,240
刚刚给大家提过了

98
00:04:06,240 --> 00:04:08,140
然后我建议大家去下载哪个模型

99
00:04:08,140 --> 00:04:09,980
去下载这一个模型

100
00:04:09,980 --> 00:04:12,440
这个mini 4B的

101
00:04:12,440 --> 00:04:13,380
你可以看到为什么

102
00:04:13,380 --> 00:04:14,520
因为这个模型

103
00:04:14,520 --> 00:04:17,280
它的整体的效果是最好的

104
00:04:17,280 --> 00:04:18,480
在这个地方你可以看到

105
00:04:18,480 --> 00:04:20,980
它不管是它精度还是速度

106
00:04:20,980 --> 00:04:22,060
它都是五颗星

107
00:04:22,060 --> 00:04:22,940
是满星的

108
00:04:22,940 --> 00:04:25,440
所以我这边是用这个模型

109
00:04:25,440 --> 00:04:26,200
给大家去演示

110
00:04:26,200 --> 00:04:27,560
然后我等一下会教大家

111
00:04:27,560 --> 00:04:28,220
怎么去下载

112
00:04:28,220 --> 00:04:28,760
对

113
00:04:28,760 --> 00:04:30,800
然后我顺便给大家提一下

114
00:04:30,800 --> 00:04:32,260
它这里有提到一个概念

115
00:04:32,260 --> 00:04:34,160
就是什么web GPU

116
00:04:34,160 --> 00:04:36,380
这个东西就是相当于是用GPU

117
00:04:36,380 --> 00:04:38,740
去做网页端的一个渲染加速的

118
00:04:38,740 --> 00:04:39,760
你网都是CPU

119
00:04:39,760 --> 00:04:41,340
所以你选它的话会更好

120
00:04:41,340 --> 00:04:42,100
速度更快

121
00:04:42,100 --> 00:04:42,920
第二步的话

122
00:04:42,920 --> 00:04:44,820
就下载机器发音的模型

123
00:04:44,820 --> 00:04:45,800
那机器发音的模型

124
00:04:45,800 --> 00:04:46,700
官方这边推荐

125
00:04:46,700 --> 00:04:47,640
我们是用这个模型

126
00:04:47,640 --> 00:04:50,640
就是这个叫做Translate Gemma

127
00:04:50,640 --> 00:04:52,580
它的速度和质量整体也是最好的

128
00:04:52,580 --> 00:04:53,940
所以我们就选它

129
00:04:53,940 --> 00:04:55,860
然后这边下载的好了之后

130
00:04:55,860 --> 00:04:58,060
这个模型大概是有将近3G左右

131
00:04:58,060 --> 00:04:58,840
然后你也看到

132
00:04:58,840 --> 00:05:00,640
它这边支持了各种各样的语言

133
00:05:00,640 --> 00:05:01,820
语种也比较丰富

134
00:05:01,820 --> 00:05:02,340
OK

135
00:05:02,340 --> 00:05:05,080
然后TTS模型就是语音合成

136
00:05:05,080 --> 00:05:06,140
这是那个波音的

137
00:05:06,140 --> 00:05:08,920
这个的话我这边是选的是这个

138
00:05:08,920 --> 00:05:09,740
Super Sonic

139
00:05:09,740 --> 00:05:11,260
这个模型我们可以看到

140
00:05:11,260 --> 00:05:12,980
它只有380多兆

141
00:05:12,980 --> 00:05:13,780
对吧

142
00:05:13,780 --> 00:05:14,480
它比较小

143
00:05:14,480 --> 00:05:16,440
然后它支持的语种也很丰富

144
00:05:16,440 --> 00:05:18,520
这样的话就是你

145
00:05:18,520 --> 00:05:21,120
组合起来就会非常的丰富了

146
00:05:21,120 --> 00:05:23,000
你语音识别跟语音合成

147
00:05:23,000 --> 00:05:24,060
包括翻译

148
00:05:24,060 --> 00:05:26,100
语种的范围就很丰富了

149
00:05:26,100 --> 00:05:27,840
你的那种可拓展性就很强了

150
00:05:27,840 --> 00:05:28,160
对吧

151
00:05:28,160 --> 00:05:28,580
OK

152
00:05:28,580 --> 00:05:29,820
那它怎么样配呢

153
00:05:29,820 --> 00:05:30,800
来我教大家

154
00:05:30,800 --> 00:05:32,600
首先我们来到语音识别这里

155
00:05:32,600 --> 00:05:33,500
你默认的话

156
00:05:33,500 --> 00:05:34,040
它是会

157
00:05:34,040 --> 00:05:35,340
这里有一个叫做

158
00:05:35,340 --> 00:05:37,000
浏览模型库

159
00:05:37,000 --> 00:05:37,900
我们点击一下

160
00:05:37,900 --> 00:05:39,680
然后这边有很多的模型

161
00:05:39,680 --> 00:05:40,300
大家可以看到

162
00:05:40,300 --> 00:05:43,100
这边我们就选择刚才已经下好的

163
00:05:43,100 --> 00:05:43,860
我这边已经下好了

164
00:05:43,860 --> 00:05:45,340
它就会显示使用中

165
00:05:45,340 --> 00:05:46,600
你没有下载好的

166
00:05:46,600 --> 00:05:47,560
我举个例子

167
00:05:47,560 --> 00:05:48,920
你就点一下下载按钮

168
00:05:48,920 --> 00:05:50,200
你点一下下载

169
00:05:50,200 --> 00:05:52,180
它就会给你开始在本地去下载了

170
00:05:52,180 --> 00:05:54,320
然后你就只要等了几分钟就好了

171
00:05:54,320 --> 00:05:56,220
这里的话需要注意的

172
00:05:56,220 --> 00:05:58,540
就是你的网络一定保持全局模式

173
00:05:58,540 --> 00:05:59,580
网络要稳定

174
00:05:59,580 --> 00:06:01,260
你看它就现在已经开始下载了

175
00:06:01,260 --> 00:06:01,500
OK

176
00:06:01,500 --> 00:06:02,960
我这边暂时不用给它下

177
00:06:02,960 --> 00:06:03,640
我就给它插掉

178
00:06:03,640 --> 00:06:05,300
我还是依然用我们的这个

179
00:06:05,300 --> 00:06:07,020
mini的4B的这个模型

180
00:06:07,020 --> 00:06:07,460
OK

181
00:06:07,460 --> 00:06:08,320
然后呢

182
00:06:08,320 --> 00:06:09,860
我们再接着去找

183
00:06:09,860 --> 00:06:11,220
刚才另外几个模型

184
00:06:11,220 --> 00:06:12,020
比如说

185
00:06:12,020 --> 00:06:13,860
语音那个翻译的模型呢

186
00:06:13,860 --> 00:06:14,700
这边也有啊

187
00:06:14,700 --> 00:06:16,740
这里我们点击这个浏览模型库

188
00:06:16,740 --> 00:06:17,620
都是一样的

189
00:06:17,620 --> 00:06:18,980
很努力操作啊

190
00:06:18,980 --> 00:06:20,640
这边我已经下载好了

191
00:06:20,640 --> 00:06:21,760
这个Translator Gemma

192
00:06:21,760 --> 00:06:22,880
刚才那个4B模型

193
00:06:22,880 --> 00:06:24,880
那其他模型你要下载也是一样的

194
00:06:24,880 --> 00:06:26,360
在这地方点下载按钮就好了

195
00:06:26,360 --> 00:06:27,120
很简单

196
00:06:27,120 --> 00:06:29,700
然后最后一个这个模型

197
00:06:29,700 --> 00:06:31,580
就是那个语音合成的这个模型

198
00:06:31,580 --> 00:06:33,480
同样还是在这里点浏览模型库

199
00:06:33,480 --> 00:06:34,660
可以看到

200
00:06:34,660 --> 00:06:35,920
这里就有很多代选的

201
00:06:35,920 --> 00:06:37,100
你根据自己的叙述来选

202
00:06:37,100 --> 00:06:38,760
它这边会有一些标签

203
00:06:38,760 --> 00:06:39,280
就是什么

204
00:06:39,280 --> 00:06:40,900
EN F AER这些

205
00:06:40,900 --> 00:06:43,420
这边是代表你语音的模型

206
00:06:43,420 --> 00:06:44,600
它支持哪些语种

207
00:06:44,600 --> 00:06:46,200
我这里选择一个很多的

208
00:06:46,200 --> 00:06:46,840
大家可以看到

209
00:06:46,840 --> 00:06:48,220
它也下载好了

210
00:06:48,220 --> 00:06:49,200
下载好之后

211
00:06:49,200 --> 00:06:50,480
这三个模型都配好了

212
00:06:50,480 --> 00:06:50,980
OK

213
00:06:50,980 --> 00:06:52,940
我们现在要把它选中

214
00:06:52,940 --> 00:06:54,500
就是你已经把它都选好了

215
00:06:54,500 --> 00:06:56,860
然后这里有语音合成的一个速度

216
00:06:56,860 --> 00:06:58,720
这个东西它默认是1.0

217
00:06:58,720 --> 00:06:59,980
我这里给它改成了1.2

218
00:06:59,980 --> 00:07:02,020
可能我觉得稍微速度会快一点

219
00:07:02,020 --> 00:07:04,020
根据个人的一个需求

220
00:07:04,020 --> 00:07:05,740
都没有一个绝对的标准

221
00:07:05,740 --> 00:07:07,380
然后底下这几个参数

222
00:07:07,380 --> 00:07:08,660
我要给大家稍微提一下

223
00:07:08,660 --> 00:07:10,460
这些都是默认已经配好的

224
00:07:10,460 --> 00:07:12,280
它都是一个比较成熟的参数

225
00:07:12,280 --> 00:07:13,800
你不用一般不用去改的

226
00:07:13,800 --> 00:07:14,800
像什么

227
00:07:14,800 --> 00:07:17,020
比如说最小的静音时长

228
00:07:17,020 --> 00:07:19,440
这一项它一般是设置为1.4秒

229
00:07:19,440 --> 00:07:20,120
是什么意思呢

230
00:07:20,120 --> 00:07:21,260
就是我们说话的时候

231
00:07:21,260 --> 00:07:23,620
如果两句话间隔超过了1.4秒的话

232
00:07:23,620 --> 00:07:26,580
那么它都会把这里切割开

233
00:07:26,580 --> 00:07:27,660
分成两句话

234
00:07:27,660 --> 00:07:28,220
就这意思

235
00:07:28,220 --> 00:07:29,780
然后最小的语音时长

236
00:07:29,780 --> 00:07:31,800
它这里默认写的是0.4秒

237
00:07:31,800 --> 00:07:32,880
你就是说

238
00:07:32,880 --> 00:07:34,440
如果你说话有一些那种

239
00:07:34,440 --> 00:07:35,840
声音比较零碎

240
00:07:35,840 --> 00:07:37,660
那个声音的片的小于0的4秒

241
00:07:37,660 --> 00:07:38,860
它都不会给你去识别

242
00:07:38,860 --> 00:07:40,000
这生的意思是杂音

243
00:07:40,000 --> 00:07:40,400
对吧

244
00:07:40,400 --> 00:07:42,120
这也是比较节省算力的一种方法

245
00:07:42,120 --> 00:07:42,720
OK

246
00:07:42,720 --> 00:07:43,720
这个都弄好之后

247
00:07:43,720 --> 00:07:45,000
我们这里还有一个音频

248
00:07:45,000 --> 00:07:46,760
这一家也是你的

249
00:07:46,760 --> 00:07:48,400
就是你的物理的外设

250
00:07:48,400 --> 00:07:49,360
我们要选什么呢

251
00:07:49,360 --> 00:07:50,140
首先你的麦克风

252
00:07:50,140 --> 00:07:52,400
你要选你的实际的物理的外设

253
00:07:52,400 --> 00:07:53,620
你不要选其他的

254
00:07:53,620 --> 00:07:54,880
不然它老实上连不通的

255
00:07:54,880 --> 00:07:56,960
然后这个东西

256
00:07:56,960 --> 00:07:58,100
其他一般就不用管

257
00:07:58,100 --> 00:08:00,140
然后还有提供商

258
00:08:00,140 --> 00:08:01,440
这个刚刚我们已经说过了

259
00:08:01,440 --> 00:08:03,540
对然后就是常规这里

260
00:08:03,540 --> 00:08:04,860
然后你要说什么

261
00:08:04,860 --> 00:08:05,600
你要说中文

262
00:08:05,600 --> 00:08:07,900
然后你希望对方听到什么

263
00:08:07,900 --> 00:08:08,620
我选英文

264
00:08:08,620 --> 00:08:09,960
那你这里的语言

265
00:08:09,960 --> 00:08:10,500
你要注意

266
00:08:10,500 --> 00:08:12,520
因为我刚才选的那些模型

267
00:08:12,520 --> 00:08:13,660
它支持了很多种语种

268
00:08:13,660 --> 00:08:15,480
所以我这里才有的这么多选

269
00:08:29,397 --> 00:08:32,457
我们目前主要是用我挣动Planet就好了

270
00:08:32,457 --> 00:08:33,457
如果是对方的话

271
00:08:33,457 --> 00:08:34,357
那就是另外一种

272
00:08:34,357 --> 00:08:35,237
挣后面的奶肉

273
00:08:35,237 --> 00:08:36,117
我们今天不演示了

274
00:08:36,117 --> 00:08:36,397
OK

275
00:08:36,397 --> 00:08:37,797
我们点击开始绘画

276
00:08:37,797 --> 00:08:39,577
这里加的好了之后

277
00:08:39,577 --> 00:08:40,857
我们现在就可以说话了

278
00:08:40,857 --> 00:08:41,697
然后你可以看到

279
00:08:41,697 --> 00:08:44,477
左上角它正在识别我们讲的话

280
00:08:44,477 --> 00:08:46,617
然后它这个是实时的

281
00:08:46,617 --> 00:08:47,537
是牛式的

282
00:08:47,537 --> 00:08:48,597
这个字有点小

283
00:08:48,597 --> 00:08:49,937
我们可以点一下右上角

284
00:08:49,937 --> 00:08:50,797
这里有一个字幕

285
00:08:50,797 --> 00:08:51,917
把它放大一些

286
00:08:51,917 --> 00:08:54,797
OK

287
00:08:54,797 --> 00:08:55,857
所以你结束了

288
00:08:55,857 --> 00:08:57,777
我们可以开始谈谈

289
00:08:57,777 --> 00:09:00,577
你会看到它在最后的角度

290
00:09:00,577 --> 00:09:02,137
它在认识我们的意思

291
00:09:02,137 --> 00:09:04,137
而且因为它在实现时

292
00:09:04,137 --> 00:09:05,397
的角度是小的

293
00:09:05,397 --> 00:09:07,877
你可以在最后的角度

294
00:09:07,877 --> 00:09:10,317
有一个下方的选择选择选择

295
00:09:10,317 --> 00:09:12,777
对 整体汉塘基特的效果还可以

296
00:09:12,777 --> 00:09:14,437
虽然说有时候可能识别的

297
00:09:14,437 --> 00:09:15,377
个别字不再精准

298
00:09:15,377 --> 00:09:17,037
当然这可能也是我个人

299
00:09:17,037 --> 00:09:18,317
有时候普通话不再标读

300
00:09:18,317 --> 00:09:19,497
所以它识别起来

301
00:09:19,497 --> 00:09:21,297
可能会有一些点点误差

302
00:09:21,297 --> 00:09:22,857
但是也不影响它的一个翻译

303
00:09:22,857 --> 00:09:24,057
我目一看到和整体

304
00:09:24,057 --> 00:09:26,217
我觉得效果还是不错的

305
00:09:27,777 --> 00:09:31,297
Overall, it seems like it performs reasonably well.

306
00:09:32,077 --> 00:09:36,177
Although sometimes it might not recognize individual characters with perfect accuracy,

307
00:09:36,657 --> 00:09:40,597
Of course, that could also be due to my occasional inconsistencies in Mandarin.

308
00:09:41,677 --> 00:09:44,677
As a result, there might be minor errors in the translation,

309
00:09:45,157 --> 00:09:47,597
But it still provides a useful translation overall,

310
00:09:47,937 --> 00:09:49,637
And I think the results are quite good.

311
00:09:51,697 --> 00:09:53,877
OK,我们退出的话,我们就点下ESC,

312
00:09:54,257 --> 00:09:55,317
然后呢,这些字幕的话,

313
00:09:55,417 --> 00:09:57,657
我们才有一个删除的小按钮啊,

314
00:09:57,657 --> 00:09:58,417
点一下

315
00:09:58,417 --> 00:10:00,017
给清空

316
00:10:00,017 --> 00:10:01,017
然后我们要结束的话

317
00:10:01,017 --> 00:10:03,297
我们点击下方结束绘画

318
00:10:03,297 --> 00:10:04,657
这就结束了

319
00:10:04,657 --> 00:10:05,297
OK

320
00:10:05,297 --> 00:10:07,417
如果你希望对方听到了

321
00:10:07,417 --> 00:10:09,497
就刚刚给大家演示到不同的语种

322
00:10:09,497 --> 00:10:10,057
这个一般

323
00:10:10,057 --> 00:10:12,217
比如说你想要他听到西班牙语

324
00:10:12,217 --> 00:10:14,217
或者是你想要他听到日本语

325
00:10:14,217 --> 00:10:15,857
你这个地方切换你就好了

326
00:10:15,857 --> 00:10:17,897
然后我们点击开始绘画

327
00:10:18,697 --> 00:10:19,257
你好啊

328
00:10:19,257 --> 00:10:20,497
可以听到我声音吗

329
00:10:24,897 --> 00:10:25,777
你好

330
00:10:25,777 --> 00:10:28,337
我想现在跟你分享一些事情

331
00:10:28,337 --> 00:10:30,537
こんにちは

332
00:10:30,537 --> 00:10:33,517
今私がお話したいことがあります

333
00:10:33,517 --> 00:10:34,557
你可以看到

334
00:10:34,557 --> 00:10:37,637
他这个字幕还是就是逐步的播放去的

335
00:10:37,637 --> 00:10:39,837
这个我觉得效果是非常好的

336
00:10:39,837 --> 00:10:42,337
就是有你讲话

337
00:10:42,337 --> 00:10:45,177
然后你可以看到这个字幕是这样流动的

338
00:10:45,177 --> 00:11:04,957
然后我们把它结束

339
00:11:04,957 --> 00:11:07,617
然后我们可以再喜欢其他的这种语言

340
00:11:07,617 --> 00:11:08,797
那这些都是类似的

341
00:11:08,797 --> 00:11:10,417
我就不给大家一一去演示了

342
00:11:10,417 --> 00:11:11,517
然后呢

343
00:11:11,517 --> 00:11:13,017
这上面有一些东西

344
00:11:13,017 --> 00:11:14,037
我再给大家讲一下

345
00:11:14,037 --> 00:11:15,437
就是提供商

346
00:11:16,597 --> 00:11:17,517
因为本地的

347
00:11:17,517 --> 00:11:19,397
其实我已经试过了这几个模型

348
00:11:19,397 --> 00:11:20,437
它都是在本地

349
00:11:20,437 --> 00:11:22,757
就没有去走任何在线的API

350
00:11:22,757 --> 00:11:23,757
如果有些小伙伴

351
00:11:23,757 --> 00:11:25,517
他说我有在线的这种API的话

352
00:11:25,517 --> 00:11:26,197
你也可以去用

353
00:11:26,357 --> 00:11:26,957
这个都行

354
00:11:26,957 --> 00:11:27,837
因为我觉得

355
00:11:27,917 --> 00:11:29,317
其实本地的效果就很棒了

356
00:11:29,477 --> 00:11:30,237
这里可以看到

357
00:11:30,677 --> 00:11:31,397
这边怎么选

358
00:11:31,597 --> 00:11:32,157
你可以看到

359
00:11:32,157 --> 00:11:33,997
首先在这个地方点常规

360
00:11:34,117 --> 00:11:36,037
然后这里有个AI服务提供商

361
00:11:36,037 --> 00:11:36,717
我们点一下

362
00:11:37,077 --> 00:11:38,317
这里有OpenAI

363
00:11:38,637 --> 00:11:39,717
有一些其他的

364
00:11:39,717 --> 00:11:40,997
比如说什么豆包

365
00:11:40,997 --> 00:11:41,997
还有谷歌

366
00:11:41,997 --> 00:11:43,677
Gemini等等的这些

367
00:11:43,677 --> 00:11:45,117
你用买哪个API

368
00:11:45,117 --> 00:11:46,757
你就给它对应的输入

369
00:11:46,757 --> 00:11:47,917
比如我点openAI的

370
00:11:47,917 --> 00:11:49,237
你看你点进openAI的

371
00:11:49,237 --> 00:11:50,557
这个地方提用你的密钥

372
00:11:50,557 --> 00:11:51,957
然后这里音频的话

373
00:11:51,957 --> 00:11:53,277
这个保守不动

374
00:11:53,277 --> 00:11:54,517
提供商这一块的话

375
00:11:54,517 --> 00:11:56,397
你非在这里选择

376
00:11:56,397 --> 00:11:58,277
就是openAI那个语音的模型

377
00:11:58,277 --> 00:12:00,637
对应的角色的发音的角色

378
00:12:00,637 --> 00:12:00,977
对吧

379
00:12:00,977 --> 00:12:01,677
你自己去选

380
00:12:01,677 --> 00:12:02,557
然后这些东西

381
00:12:02,557 --> 00:12:04,517
你可以把它参数对应的填过去

382
00:12:04,517 --> 00:12:06,477
它就可以调在线的API了

383
00:12:06,477 --> 00:12:06,757
对

384
00:12:06,757 --> 00:12:07,997
原理是一样的

385
00:12:07,997 --> 00:12:09,357
但是我自己是觉得

386
00:12:09,357 --> 00:12:11,257
其实我们免费的

387
00:12:11,257 --> 00:12:12,437
我今天给大家展示的

388
00:12:12,437 --> 00:12:13,177
主要是免费的

389
00:12:13,177 --> 00:12:14,517
你已经够用了

390
00:12:14,517 --> 00:12:14,937
对

391
00:12:14,937 --> 00:12:17,617
所以整体实际下来这个效果是

392
00:12:17,617 --> 00:12:19,177
我觉得是在我目前

393
00:12:19,177 --> 00:12:21,057
就是看过的所有开源的这种

394
00:12:21,057 --> 00:12:22,057
童真传映的项目里面

395
00:12:22,057 --> 00:12:22,917
应该是最好的了

396
00:12:22,917 --> 00:12:24,217
所以我强烈建议大家

397
00:12:24,217 --> 00:12:25,997
尤其是很多小伙伴经常需要

398
00:12:25,997 --> 00:12:28,737
用到外语去交流

399
00:12:28,737 --> 00:12:29,997
但是自己可能外语水平

400
00:12:29,997 --> 00:12:30,897
可能还没那么好

401
00:12:30,897 --> 00:12:31,897
用这些工具

402
00:12:31,897 --> 00:12:33,457
我觉得是对你有很大的帮助的

403
00:12:33,457 --> 00:12:33,777
OK

404
00:12:33,777 --> 00:12:35,697
本期的这些课程资源

405
00:12:35,697 --> 00:12:36,377
其实比较少

406
00:12:36,377 --> 00:12:37,677
但是我会放到评论区

407
00:12:37,677 --> 00:12:38,677
大家直接可以去用

408
00:12:38,677 --> 00:12:40,817
我最后还要提到一个注意点

409
00:12:40,817 --> 00:12:42,517
就是我在实际使用步骤中

410
00:12:42,517 --> 00:12:44,877
很多小伙伴使用的时候会报错

411
00:12:44,877 --> 00:12:46,977
这个报错是什么原因造成的

412
00:12:46,977 --> 00:12:48,937
我们一定要注意

413
00:12:48,937 --> 00:12:50,757
我们用到的这三个模型

414
00:12:50,757 --> 00:12:53,077
它支持的语语种范围是什么

415
00:12:53,077 --> 00:12:54,377
举个例子

416
00:12:54,377 --> 00:12:56,077
比如说我这里的语语种

417
00:12:56,077 --> 00:12:57,317
它有一个粤语

418
00:12:57,317 --> 00:12:58,377
这也是支持的

419
00:12:58,377 --> 00:12:59,597
我们可以看到这有粤语

420
00:12:59,597 --> 00:13:02,257
但是我这里说话的话

421
00:13:02,257 --> 00:13:03,437
我说的是普通话

422
00:13:03,437 --> 00:13:04,957
然后我语音识别

423
00:13:04,957 --> 00:13:06,237
可能它这个模型

424
00:13:06,237 --> 00:13:07,697
它能够识别普通话

425
00:13:07,697 --> 00:13:10,117
但是它语音合成的模型

426
00:13:10,117 --> 00:13:11,837
它不支持播放粤语

427
00:13:11,837 --> 00:13:12,877
那么它就会报错

428
00:13:12,877 --> 00:13:13,957
你的三个模型

429
00:13:13,957 --> 00:13:16,777
大家可以看到吧

430
00:13:16,777 --> 00:13:17,937
就是你的三个模型

431
00:13:17,937 --> 00:13:19,237
它们支持的与种

432
00:13:19,237 --> 00:13:20,617
一定要有共同的

433
00:13:20,617 --> 00:13:22,917
那么这个念路才是通的

434
00:13:22,917 --> 00:13:24,717
有一个不支持的

435
00:13:24,717 --> 00:13:25,997
那么你这个念路就通不了

436
00:13:25,997 --> 00:13:27,677
小伙伴可能遇到的这些

437
00:13:27,677 --> 00:13:28,557
下部分可能是

438
00:13:28,557 --> 00:13:29,717
这一方面的一个问题

439
00:13:29,717 --> 00:13:31,317
还有就是你网络不好

440
00:13:31,317 --> 00:13:32,897
你网络如果有问题的话

441
00:13:32,897 --> 00:13:34,157
你没开全局模式的话

442
00:13:34,157 --> 00:13:35,997
你这些模型是下载不到你本地的

443
00:13:35,997 --> 00:13:36,877
还有一个细节就是

444
00:13:36,877 --> 00:13:37,877
关明这里有个登录

445
00:13:37,877 --> 00:13:39,497
这里刚才跟大家有讲过

446
00:13:39,497 --> 00:13:40,817
你看这个登录这里

447
00:13:40,817 --> 00:13:42,057
你可以点击去注册

448
00:13:42,057 --> 00:13:44,537
或者是就已经有账号就登陆进去

449
00:13:44,537 --> 00:13:45,677
那它本质上呢

450
00:13:45,677 --> 00:13:47,777
它会用到它云端的这个Sirculture

451
00:13:47,777 --> 00:13:49,317
它自己的翻译服务

452
00:13:49,317 --> 00:13:51,737
那对于有些小伙伴不想去折腾

453
00:13:51,737 --> 00:13:52,697
你可以去用它

454
00:13:52,697 --> 00:13:53,417
那这样的话

455
00:13:53,417 --> 00:13:55,177
也不用去本地下载模型了

456
00:13:55,177 --> 00:13:55,657
但是呢

457
00:13:55,657 --> 00:13:56,797
要提醒一下大家就是

458
00:13:56,797 --> 00:13:57,597
你这样的话

459
00:13:57,597 --> 00:13:59,457
你的数据是会经过它的服务器的

460
00:13:59,457 --> 00:14:00,977
就是说数据的隐私性

461
00:14:00,977 --> 00:14:01,857
找找没那么好嘛

462
00:14:01,857 --> 00:14:02,737
根据自己的叙手来

463
00:14:02,737 --> 00:14:04,057
如果你全都想让数据

464
00:14:04,057 --> 00:14:05,497
不出自己的本地的电脑

465
00:14:05,497 --> 00:14:07,357
那么你就用我刚刚给大家讲的方大

466
00:14:07,357 --> 00:14:08,257
还有一点呢

467
00:14:08,257 --> 00:14:09,677
也是我个人使用的一个经验

468
00:14:09,677 --> 00:14:11,977
这个东西对大家应该很有帮助

469
00:14:11,977 --> 00:14:13,557
同声传译这个东西

470
00:14:13,557 --> 00:14:14,097
首先呢

471
00:14:14,097 --> 00:14:15,737
大家要知道它最看重的是什么

472
00:14:15,737 --> 00:14:17,637
就是一个是翻译的准确性

473
00:14:17,637 --> 00:14:19,417
一个是它的一个实言

474
00:14:19,417 --> 00:14:21,697
就是我们那个翻译的那个时间

475
00:14:21,697 --> 00:14:22,517
不能太久了

476
00:14:22,517 --> 00:14:24,217
这样的话就做不到同声了

477
00:14:24,217 --> 00:14:26,717
那我们要想在这个项目中

478
00:14:26,717 --> 00:14:29,457
要尽量的去把它实言占低

479
00:14:29,457 --> 00:14:31,037
让大家的用户体验更好

480
00:14:31,037 --> 00:14:32,197
应该怎么去做呢

481
00:14:32,197 --> 00:14:33,997
首先刚才我跟大家是讲了

482
00:14:33,997 --> 00:14:35,237
它里面有很多模型

483
00:14:35,237 --> 00:14:35,577
对吧

484
00:14:35,577 --> 00:14:37,797
大家可以看到这模型是有一大堆的

485
00:14:37,797 --> 00:14:40,137
那么我们应该怎么去选模型的

486
00:14:40,137 --> 00:14:40,837
大家要记住

487
00:14:40,837 --> 00:14:41,797
你要选

488
00:14:41,797 --> 00:14:43,077
可能要去选牛式的

489
00:14:43,077 --> 00:14:44,077
也就是说

490
00:14:44,077 --> 00:14:45,437
不要选非牛式的模型

491
00:14:45,437 --> 00:14:46,597
牛式的模型一般

492
00:14:46,597 --> 00:14:47,297
会带上一个叫

493
00:14:47,297 --> 00:14:48,477
real time的字段

494
00:14:48,477 --> 00:14:51,537
然后有webGPU的

495
00:14:51,537 --> 00:14:52,757
你就进来就选它

496
00:14:52,757 --> 00:14:53,117
对

497
00:14:53,117 --> 00:14:54,897
这个是一个细节

498
00:14:54,897 --> 00:14:55,517
很重要的

499
00:14:55,517 --> 00:14:57,197
就是你选那种非牛式的

500
00:14:57,197 --> 00:14:57,977
它推你的话

501
00:14:57,977 --> 00:14:59,377
那整个内容就会很慢了

502
00:14:59,377 --> 00:15:00,497
所以等于说

503
00:15:00,497 --> 00:15:01,797
你说完半句话以后

504
00:15:01,797 --> 00:15:03,437
别人都已经说完

505
00:15:03,437 --> 00:15:04,017
下一句了

506
00:15:04,017 --> 00:15:05,077
这还没翻译完

507
00:15:05,077 --> 00:15:06,757
当然那体验就很不好

508
00:15:06,757 --> 00:15:07,157
对吧

509
00:15:07,157 --> 00:15:08,617
所以大家一定要记住

510
00:15:08,617 --> 00:15:10,297
这是要近代选牛逝的模型

511
00:15:10,297 --> 00:15:11,837
不管是语音识别

512
00:15:11,837 --> 00:15:12,957
还是翻译

513
00:15:12,957 --> 00:15:14,377
或者是语音合成

514
00:15:14,377 --> 00:15:15,437
都要去选牛逝的

515
00:15:15,437 --> 00:15:16,837
然后有GPU的话

516
00:15:16,837 --> 00:15:19,717
就要去把webGPU也选好

517
00:15:19,717 --> 00:15:21,977
本期视频我就跟大家分享到这了

518
00:15:21,977 --> 00:15:23,177
大家如果觉得我的视频

519
00:15:23,177 --> 00:15:24,057
对你有帮助的话

520
00:15:24,057 --> 00:15:25,877
也请支持一下我的创作

521
00:15:25,877 --> 00:15:27,277
帮我点赞 关注 转发

522
00:15:27,277 --> 00:15:28,917
我们本期内容到此结束

523
00:15:28,917 --> 00:15:29,857
我们下期再会