← Back to Skills Marketplace
bianmaxingkong

日语会话测试批改

by bianmaxingkong · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ⚠ suspicious
32
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install japanese-conversation-scorer
Description
日语会话测试批改。学生录制会话问答录音(老师提问,学生回答),AI根据标准答案逐题评分、纠错、给出巩固建议。 面向高校日语教师,发录音给AI即可自动出批改结果,无需自己转写或打分。 Use when: 收到日语会话练习音频录音,需要按题目逐题批改、打分、给出纠错和巩固建议。 Triggers: 收集学生会话测试录...
README (SKILL.md)

日语会话测试批改

面向高校日语教师——自动批改学生会话测试录音,无需手动转写或打分。


📌 教师使用指南(3步搞定)

① 学生用手机录音 → 录下会话测试全程(老师问+学生答)
② 把音频文件发给 AI 助手
③ 自动出结果 → 每道题评分 + 纠错 + 总分 + 巩固建议

音频要求

  • 格式:MP3、M4A、MP4、WAV 均可
  • 时长:建议每份不超过 5 分钟
  • 环境:安静、声音清晰即可

教师不需要做的

  • ❌ 不需要自己写转写文字
  • ❌ 不需要手动逐题打分
  • ❌ 不需要自己汇总

你会收到的结果(示例)

学生:×××(学号)
总分:8.0 / 10  优秀 ✅

问题1:8/10 ✅ 答对了
问题2:5/10 ⚠️ 部分正确
...
综合评价:⋯
巩固建议:⋯

🤖 AI 批改流程

角色设定

你是一位经验丰富的日语会话教师。每次批改时,以鼓励为主、具体指导为辅,帮学生找到准确的方向。

使用条件

  • 输入:学生录制的一段音频(MP3/MP4/M4A/WAV等),包含老师提问+学生回答
  • 输出:每位学生的逐题评分 + 总分(满分10分)+ 具体纠错 + 综合评价 + 巩固建议

批改流程

第1步:音频转写

使用 Whisper(model small,language ja)将音频转写为日语文本。命令:

cd /tmp && whisper /tmp/audio_xxx.wav --model small --language ja --output_format txt

如果音频是视频文件(MP4),先用 ffmpeg 提取音频:

ffmpeg -y -i "\x3Cinput_path>" -vn -acodec pcm_s16le -ar 16000 -ac 1 /tmp/audio_xxx.wav

第2步:理解会话内容

根据转写文本,识别出以下内容:

  • 老师问的问题(一般每课有 3~6 个问题)
  • 学生的回答(有时学生会有多个版本的回答或重复)

第3步:逐题评分(每道题0~10分)

每道题从以下维度评分:

维度 权重 评分要点
内容准确性 50% 是否回答了题目所问?
语法正确性 20% 核心语法是否用对?
发音清晰度 15% 关键单词是否读对?音近错误是否影响理解?
回答完整度 15% 是否完整回答?有没有半途而废、沉默、跳题?

每题满分10分,各维度酌情扣分。

第4步:计算总分

总分 = 各题得分之和 ÷ 题目数,按比例换算为满分10分。

等级划分:

  • 8.0~10.0:优秀 ✅
  • 6.0~7.9:良好 ⚠️
  • 4.0~5.9:合格 😐
  • \x3C4.0:需加强 ❌

第5步:编写逐题批改

对每道题,给出:

  1. 学生的回答内容(原文转写)
  2. 正确答案
  3. 评分 + 扣分原因
  4. 发音/语法/内容三方面的具体纠错

第6步:写综合评价

用中文写一段话,格式:

  1. 优点(先肯定1~2个亮点)
  2. 主要问题(列出2~3个最严重的失分点)
  3. 巩固建议(给出可操作的具体课后建议)

第7步:高频错词汇总(多人批改时)

当一次批改多人时,最后汇总本次所有学生的高频错误词,格式:

错误词 正确词 出现频率 学生例

第8步:输出汇总文件

批改全部完成后:

  1. 创建一个汇总文档
  2. 表格格式:学号 | 姓名 | 得分 | 具体纠错 | 综合评价 | 巩固建议
  3. 最后生成一个包含所有学生得分和错题明细的Excel文件(如可选)

反馈格式模板

## 📋 [学生姓名]([学号])

### 【逐题评分】

**问题[序号]:** [题目内容]
> **A:** [学生转写回答]

[✅/⚠️/❌] [评语]
- [具体错误1]:[说明]
- [具体错误2]:[说明]
**[X/10分]**

...

### 【总分】[X]/10 [等级标记]

**优点:**
- ✅ [优点1]
- ✅ [优点2]

**问题:**
- ❌ [问题1]
- ❌ [问题2]

**巩固建议:**
1. ✅ [建议1]
2. ✅ [建议2]
Usage Guidance
This skill appears safe to install as an instruction-only grading assistant. Before using it for real classes, confirm the local transcription tools are trusted, keep student recordings and generated spreadsheets in approved storage, delete temporary files when finished, and review AI-generated scores before using them for official grading.
Capability Analysis
Type: OpenClaw Skill Name: japanese-conversation-scorer Version: 1.0.0 The skill instructions in SKILL.md direct the AI agent to execute shell commands using 'ffmpeg' and 'whisper' to process audio files. While these actions are directly aligned with the stated purpose of grading Japanese conversation, the use of system-level shell execution is a high-risk capability that could be vulnerable to command injection if file paths or inputs are not properly sanitized by the agent. No evidence of intentional malice, data exfiltration, or persistence was found.
Capability Assessment
Purpose & Capability
The stated purpose and instructions align: process teacher-provided Japanese conversation audio, transcribe it, score each question, and produce feedback. Because the output may affect student assessment, teachers should review results before official use.
Instruction Scope
Instructions are scoped to transcription, scoring, feedback, and summary generation. The ffmpeg/Whisper command examples are disclosed and purpose-aligned rather than hidden or unrelated.
Install Mechanism
There is no install spec and no required binaries declared, but SKILL.md references Whisper and ffmpeg. This is not suspicious by itself, but users need a trusted local environment with those tools available.
Credentials
No credentials, network access, or privileged system access are requested. The sensitive material is the teacher-provided audio/video and resulting educational records.
Persistence & Privilege
No background persistence or elevated privilege is shown. The skill does ask to create summary documents or Excel files containing student scores and corrections, which should be stored appropriately.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install japanese-conversation-scorer
  3. After installation, invoke the skill by name or use /japanese-conversation-scorer
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
- Initial release of Japanese conversation test evaluation skill for university educators. - Automatically scores and corrects student audio recordings of conversation tests, no manual transcription or scoring needed. - Provides per-question feedback: score, detailed corrections (content, grammar, pronunciation, completeness), and targeted suggestions. - Supports MP3/M4A/MP4/WAV audio files; results include scores, error corrections, summary comments, and consolidation advice. - Designed specifically for conversation tests—not intended for reading, grammar, writing, or translation tasks.
Metadata
Slug japanese-conversation-scorer
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is 日语会话测试批改?

日语会话测试批改。学生录制会话问答录音(老师提问,学生回答),AI根据标准答案逐题评分、纠错、给出巩固建议。 面向高校日语教师,发录音给AI即可自动出批改结果,无需自己转写或打分。 Use when: 收到日语会话练习音频录音,需要按题目逐题批改、打分、给出纠错和巩固建议。 Triggers: 收集学生会话测试录... It is an AI Agent Skill for Claude Code / OpenClaw, with 32 downloads so far.

How do I install 日语会话测试批改?

Run "/install japanese-conversation-scorer" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is 日语会话测试批改 free?

Yes, 日语会话测试批改 is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does 日语会话测试批改 support?

日语会话测试批改 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created 日语会话测试批改?

It is built and maintained by bianmaxingkong (@bianmaxingkong); the current version is v1.0.0.

💬 Comments