我如何编写 ElevenLabs 可以实际执行的播客介绍
人工智能旁白在标点符号上失败,而不是在单词上。 These five writing rules turned my robotic-sounding intros into ones listeners assume are recorded.
我的第一个 ElevenLabs 介绍听起来像是航班广播,问题从来不是声音,而是我为读者而不是表演者而写。 Five writing rules fixed it, and they take five minutes to apply.
规则一:将标点符号写为表现。 A comma is a breath, a period is a full stop, an em dash is a pickup in energy. Rule two: short sentences, always under fifteen words for emphasis lines. Long sentences produce flat AI delivery because the model spreads its attention evenly.
Rule three: spell numbers and abbreviations the way they should sound, write twenty-twenty-six and N-A-S-A.规则四:谨慎而谨慎地使用省略号,每个段落最多使用一个省略号,因为每个省略号都会导致戏剧性的停顿,而货币会快速膨胀。
第五条规则是区分体面与良好的一条:通过重组句子来标记每个句子中重要的一个单词,以便该单词放在最后。 Models handle emphasis through position better than through caps or quotes, which mostly produce weird stress artifacts.
将所有五个应用到九十秒的前奏中,生成稳定性略低于默认值的两到三个镜头,然后选择可以通过编辑一个单词来修复错误的镜头。我的听众调查无法看出它是合成的,更新工作流程、编辑文本、重新生成一个段落,意味着我的介绍每周都会更改,而不是像真实录音那样每年更改一次。
