#5273 字幕存在断句错误情况 上下语句不连贯

2a09:bac1* Posted at: 2 days ago

1
00:00:00,000 --> 00:00:09,360
Wet work is the old intelligence community term for assassination operations. The name comes from the fact that these missions often leave behind blood; hence, the wet part. A series.

2
00:00:09,360 --> 00:00:18,576
These increasingly bold assassinations have now struck Russia's high command far from the front lines, and what most people don't realize is that all of these killings are part of

3
00:00:18,576 --> 00:00:27,792
a much larger wet work campaign on a scale that we haven't seen since World War Two, in a campaign that's designed to pressure Russian authorities into ending

4
00:00:27,792 --> 00:00:37,008
the war, according to leaked European intelligence reports, Putin has now become so fearful of these operations that he has stayed in his underground bunker for weeks on end

5
00:00:37,008 --> 00:00:46,224
to avoid these assassination attempts. So today I want to look at the unfolding assassination campaign, and we'll ask if this kind of assassination

6
00:00:46,224 --> 00:00:55,440
strategy like this even works for nearly three years. One of Russia's most senior generals lived with the kind of security most people can only imagine. He had a lead

1 Replies

这是 ASR 语音识别后的断句不合理导致的,主要表现为单条字幕过长或语义切点错误。可以用以下两种方式改善:

  1. 开启 LLM 重新断句
    在主界面第 2 行语音识别渠道旁,选择“LLM 重新断句”,并确保已配置 DeepSeek 或 OpenAI 等 AI 渠道的 API Key。系统会将识别文本发送给 AI,自动修正错别字并重新切分句子,使上下字幕语义连贯。
  2. 调整语音识别参数
    进入 菜单 → 工具 → 高级选项 → 语音识别参数,减小“最长语音持续秒数”(例如设为 6~8),增大“最短语音持续毫秒”(例如设为 3000),并勾选“合并过短字幕到邻近”,能让字幕更好地按自然语流分割。

注意:如果使用了clone音色克隆,不要开启 LLM 重新断句,否则会破坏参考音频对齐。

请查阅相关文档:

Post Your Reply
Open source and free maintenance is not easy. If this project is helpful to you, please consider making a small donation to help the project continue to maintain and update. Dondate: ko-fi.com/jianchang512

Related resource