by syydaniel
开口 kaikou:让克隆的声音,一开口就说人话。A skill for AI coding assistants: natural-sounding voice-overs from open-source voice cloning, from recording to final mix.
# Add to your Claude Code skills
git clone https://github.com/syydaniel/kaikouGuides for using ai agents skills like kaikou.
See how kaikou compares with popular alternatives.
kaikou is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by syydaniel. 开口 kaikou:让克隆的声音,一开口就说人话。A skill for AI coding assistants: natural-sounding voice-overs from open-source voice cloning, from recording to final mix. It has 50 GitHub stars.
kaikou's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/syydaniel/kaikou" and add it to your Claude Code skills directory (see the Installation section above). kaikou ships a SKILL.md manifest, so compatible agents can discover and load it automatically.
kaikou is primarily written in Python. It is open-source under syydaniel on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh kaikou against similar tools.
No comments yet. Be the first to share your thoughts!
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM lists this project for discovery and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
开源的克隆模型已经很强了,可同一个人、同一个模型,做出来的效果可以相差非常远。 差距在流程里:录音的环境、参考的讲法、挑哪一遍、稿子怎么写、后期怎么收。AI 味不光藏在字里,还藏在声音里,光改字只改对了一头。
这份说明给你(AI)看。每一步的细节在 docs/ 里,用到哪步读哪步。
下文的 tools/、docs/ 都在这个 skill 自己的目录里(和本文件同一层),用的时候换成完整路径。
| 步 | 做什么 | 产出 | 细节 |
|---|---|---|---|
| 1 录音与预处理 | 先空录 5 秒量底噪;环境不行先换环境;清理参考 | ref/room.wav、ref/*_clean.wav |
docs/01_录音与预处理.md |
| 2 分场景录参考 | 讲解、聊天、讲故事、中英夹杂、有劲的开场,每种 3 遍 | ref/<场景>/ |
docs/02_分场景参考.md |
| ⏸ 停一 | 用一段没念过的稿子,每种参考各配一遍,让用户盲听选(tools/audition/) |
ref/audition/round1/picks.json |
tools/audition/README.md |
| 3 克隆 | 一句一句生成,每句 4 遍 | vo/takes/ |
docs/03_克隆.md |
| 4 挑句 | 先看念对,再看像不像,最后看和前后句协不协调 | vo/lines/ |
docs/04_挑句.md |
| ⏸ 停二 | 让用户试听一段(30 到 60 秒),确认声音 | vo/sample.wav |
|
| 5 改文案 | 配出来听,哪句一听就是 AI,就改哪句,只重配改过的句子 | script.txt 新版 |
docs/05_改文案.md |
| ⏸ 停三 | 改过的稿子和重配的句子给用户看、听 | ||
| 6 后期 | 变速不变调、停顿规整、逐句响度拉平、轻处理 | vo/narration.wav |
docs/06_后期.md |
| 7 音效设计(可选) | 空谷回响、老式录音机、水下说话 | vo/fx/ |
docs/07_音效设计.md |
| ⏸ 停四 | 整条旁白和检查报告一起交给用户 |
用户已经有稿子、只要配音,就从第 1 步做到第 6 步,第 5 步只查不大改。 用户只要改稿子(不配音),就只做第 5 步,但要告诉他:稿子的毛病很多要配出来才听得见。 用户明确说"不用停"时可以跳过停顿,但第 4 步的回听和第 6 步的响度检查一定要跑。
python tools/doctor.py,看缺什么。三个克隆模型目前在苹果芯片的 Mac 上用 mlx-audio 实测过;别的平台按它给的替代办法走,并告诉用户。tools/recorder/:浏览器打开,按提示一条一条录,页面会实时显示底噪和信噪比。项目/
script.txt 稿子(格式和《卡字》一样:一句一行,| 后面是停顿秒数)
ref/ 房间声、各场景参考、清理后的参考、试听
vo/ takes/ 每句所有遍,lines/ 每句选中的那一遍,narration.wav 整条
bench/ 实测记录
录音
参考
克隆
挑句
改文案
python tools/lint_script.py script.txt --idioms idioms.txt。后期
tools/tempo.py),不要重新生成。tools/pace.py),只动静音;逐句统一响度,句间响度差控制在 0.5 LU 以内。| 工具 | 做什么 | 来源 |
|---|---|---|
tools/recorder/ |
浏览器录音页:逐条提示、实时底噪和信噪比、录完回放 | 《AI 做视频工作流》那期的录音页 |
tools/audition/ |
盲听页:候选打乱、藏起名字、响度拉平,两项打分,选定后才揭晓 | 《AI 做视频工作流》那期的盲听页,重写 |
tools/ref_prep.py |
参考清理:去直流、70 Hz 高通、按需降噪、WPE 去混响、裁尾 | 《卡字》,加了 WPE 和裁尾 |
tools/voice.py |
一句一句生成、每句 4 遍、回听打分 | 《卡字》,三个模型的接法见 docs/03 |
tools/polyphone.py |
多音字检查 | 《卡字》 |
tools/lint_script.py |
稿子检查(人话规矩) | 《卡字》,加了 AI 稿高频写法、句长变化、语气词、口头禅、成语 |
tools/pace.py、tools/tempo.py、tools/timeline.py |
停顿规整、变速不变调、逐字时间 | 《卡字》 |
tools/fx.py |
人声音效预设 | 开口新写 |
tools/doctor.py |
环境检查 | 《卡字》 |
让克隆的声音,一开口就说人话。
开源的声音克隆模型已经很强了,可同一个人、同一个模型,做出来的效果可以相差非常远。AI 味不光藏在字里,还藏在声音里:录音的环境、参考的讲法、挑哪一遍、稿子怎么写、后期怎么收,每一步都会留下痕迹。
开口是一个给 AI 助手(Claude Code、Codex 等)用的 skill,把这一整条流程交给它:
和只改稿子的去 AI 味工具不同,开口管的是从录音到成片的整条链。
把这个仓库的地址发给你的 AI 助手,跟它说"帮我安装这个 skill"。或者手动:
git clone https://github.com/syydaniel/kaikou ~/.claude/skills/kaikou
pip install -r ~/.claude/skills/kaikou/tools/requirements.txt
python ~/.claude/skills/kaikou/tools/doctor.py
然后跟 AI 说:"用开口,帮我用自己的声音配这段旁白。"
tools/doctor.py 给的替代办法)。SKILL.md 给 AI 看的总说明
docs/ 每一步的规矩和原因
tools/ 录音页、参考清理、克隆、回听打分、稿子检查、停顿和变速、音效
开口是《卡字》(让讲解动画卡在旁白的字上,https://github.com/syydaniel/kazi)的姊妹 skill。卡字管"画面跟着声音走",开口管"声音本身像个人"。两个可以一起用:先用开口配出旁白,再交给卡字做画面。
MIT。三个克隆模型各自的许可:Qwen3-TTS 和 VoxCPM 是 Apache-2.0,LongCat-AudioDiT 是 MIT,都允许商用;用之前请再看一眼它们的最新许可。克隆别人的声音之前,请先取得本人同意。