Audio Transcription Report【免费下载链接】agentic-awesome-skillsAAS Core is the local, agent-first control plane for complete catalog discovery, agent-owned selection, stack validation, and planning, backed by 2,600+ agentic skills. Includes CLI, local MCP, catalog, plugins, and Workbench.项目地址: https://gitcode.com/GitHub_Trending/an/agentic-awesome-skills📊 MetadataFieldValueFile Name{filename}File Size{file_size}Duration{duration_hms}Language{language} ({language_code})Processed Date{process_date}Speakers Identified{num_speakers}Transcription Engine{engine} (model: {model})📋 Meeting MinutesParticipants{speaker_1}{speaker_2}...Topics Discussed{topic_1}({timestamp}){key_point_1}{key_point_2}{topic_2}({timestamp}){key_point_1}Decisions Made✅ {decision_1}✅ {decision_2}Action Items{action_1}- Assigned to: {speaker} - Due: {date_if_mentioned}{action_2}- Assigned to: {speaker}Generated by audio-transcriber skill v1.0.0Transcription engine: {engine} | Processing time: {elapsed_time}s### 完整转写段落样式 README 给出了真实形态的输出样例:全文转写部分以 `**[00:00:12 → 00:00:45]** *Speaker 1*` 形式按说话人分段渲染,随后是会议纪要(Participants / Topics Discussed / Decisions Made / Action Items)与 Executive Summary(含 Key Points 与 Next Steps)。 Python 实现中的原始转写文件(`transcript-*.md`)则采用如下格式,时间戳由分段 `start`/`end` 计算为 `MM:SS`: ```markdown # Transcrição de Áudio **Arquivo:** meeting.mp3 **Idioma:** PT **Data:** 2026-02-03 02:30:45 --- ## Transcrição Completa **[00:12 → 00:45]** Good morning everyone. Let's start with updates from the frontend team. **[00:46 → 01:23]** We completed the dashboard redesign and deployed to staging yesterday.文件名与防覆盖机制v1.1.0 起改用时间戳命名,彻底解决重复运行覆盖问题(CHANGELOG 中记录为修复项 #5):TIMESTAMP=$(date +%Y%m%d-%H%M%S) TRANSCRIPT_FILE="transcript-${TIMESTAMP}.md" ATA_FILE="ata-${TIMESTAMP}.md"Python 侧save_outputs()用datetime.now().strftime("%Y%m%d-%H%M%S")生成时间戳,始终先保存transcript-*.md,若 LLM 生成了纪要则额外保存ata-*.md,且文件写入均通过safe_user_path()校验路径不越出当前工作区(相对路径、expanduser展开后必须以Path.cwd()为前缀,否则抛出ValueError),从源码层面杜绝了路径穿越风险。LLM 智能摘要工作流这是 v1.1.0 的核心升级。技能通过detect_cli_tool()按优先级探测 LLM CLI:claude存在则优先,其次检测gh copilot是否可用,两者皆无则退化为纯转写模式(仅输出 transcript,并在结尾提示安装方式)。Scenario A:用户提供自定义提示词流程(详细指南 Step 3 与 transcribe.py 的handle_prompt_workflow()对应):以 Panel 展示用户原始提示词;若检测到~/.copilot/skills/prompt-engineer/SKILL.md存在,自动调用 prompt-engineer 优化提示词(不询问,直接改进);并排展示"优化版"与"原版"两份提示词;询问💡 Usar versão melhorada? [s/n],默认s;按所选版本继续处理。若 prompt-engineer 不可用,则直接使用用户原始提示词。Scenario B:无提示词时自动生成询问"Gerar prompt automaticamente?"(默认s,选n则跳过 LLM、仅输出 transcript);构造 meta-prompt 分析转写文本(取前 4000 字符),要求用 2-3 行建议:内容类型(会议/讲座/访谈)、推荐输出格式(正式纪要/执行摘要/结构化笔记)、推荐框架(RISEN/RODES/STAR 等);展示建议并确认格式;再让 prompt-engineer 基于建议生成完整结构化提示词;预览生成的提示词并最终确认;任何一步被拒绝都回退到内置的DEFAULT_MEETING_PROMPT。内置回退提示词(DEFAULT_MEETING_PROMPT)是一个完整的 RISEN 风格模板,指定了角色(专业文档化转写员)、指令(识别内容类型、提取主题、识别说话人、提取决策与行动项、组织 Markdown)、步骤与收窄规则(保持客观清晰、保留重要上下文、使用合适的 Markdown、必要时包含时间戳),保证无任何外部依赖时也能产出高质量纪要。LLM 调用与超时处理process_with_llm()把prompt + "\n\n---\n\nTranscrição:\n\n" + transcript拼为完整输入:Claude 路径:subprocess.run(['claude', '-'], input=full_prompt, ...),即通过 stdin 管道喂给 Claude CLI;Copilot 路径:['gh', 'copilot', 'suggest', '-t', 'shell', full_prompt];两者均设 300 秒(5 分钟)超时,失败或超时时打印错误摘要并返回None,主流程随即降级为仅保存 transcript,保证任务不中断。转写与 LLM 处理全程有可视化反馈:转写分段用tqdm进度条,LLM 处理用rich的SpinnerColumn转圈动画,状态消息按颜色区分(绿=成功、黄=警告、红=错误)。实际使用:触发短语与三种典型场景SKILL.md 与 README 共同定义了激活短语:"transcribe audio to markdown"、"transcreva este áudio"、"convert audio file to text"、"extract speech from audio"、"áudio para texto com metadados"(技能原生支持葡萄牙语指令,与示例中的 pt-BR 语言检测相呼应)。场景一:基础转写copilot> transcribe audio to markdown: meeting-2026-02-02.mp3SKILL.md 展示的完整输出流为:引擎与 ffmpeg 检测 → 文件元数据(大小 12.3 MB、时长 00:45:32)→ 处理进度条 → 语言检测(pt-BR)→ 识别 4 位说话人 → 生成 Markdown,最终产出meeting-2026-02-02.md,报告 6842 词、耗时 127s,并给出下一步建议(审阅纪要、分享报告、跟踪行动项)。场景二:批处理copilot> transcreva estes áudios: recordings/*.mp3SKILL.md 的批处理输出显示:先枚举目录内 5 个文件(team-standup、client-call、brainstorm-session、product-demo、retrospective),逐个转写并报告各自耗时,最终汇总"Generated 5 Markdown reports / Total processing time"。批处理同样支持限定类型,如transcribe audio: recordings/*.wav只处理 WAV 文件。场景三:大文件预警与确认copilot> transcribe audio to markdown: conference-keynote.mp3【免费下载链接】agentic-awesome-skillsAAS Core is the local, agent-first control plane for complete catalog discovery, agent-owned selection, stack validation, and planning, backed by 2,600+ agentic skills. Includes CLI, local MCP, catalog, plugins, and Workbench.项目地址: https://gitcode.com/GitHub_Trending/an/agentic-awesome-skills创作声明:本文部分内容由AI辅助生成(AIGC),仅供参考
Audio Transcription Report【免费下载链接】agentic-awesome-skillsAAS Core is the local, agent-first control plane for complete catalog discovery, agent-owned selection, stack validation, and planning, backed by 2,600+ agentic skills. Includes CLI, local MCP, catalog, plugins, and Workbench.项目地址: https://gitcode.com/GitHub_Trending/an/agentic-awesome-skills📊 MetadataFieldValueFile Name{filename}File Size{file_size}Duration{duration_hms}Language{language} ({language_code})Processed Date{process_date}Speakers Identified{num_speakers}Transcription Engine{engine} (model: {model})📋 Meeting MinutesParticipants{speaker_1}{speaker_2}...Topics Discussed{topic_1}({timestamp}){key_point_1}{key_point_2}{topic_2}({timestamp}){key_point_1}Decisions Made✅ {decision_1}✅ {decision_2}Action Items{action_1}- Assigned to: {speaker} - Due: {date_if_mentioned}{action_2}- Assigned to: {speaker}Generated by audio-transcriber skill v1.0.0Transcription engine: {engine} | Processing time: {elapsed_time}s### 完整转写段落样式 README 给出了真实形态的输出样例:全文转写部分以 `**[00:00:12 → 00:00:45]** *Speaker 1*` 形式按说话人分段渲染,随后是会议纪要(Participants / Topics Discussed / Decisions Made / Action Items)与 Executive Summary(含 Key Points 与 Next Steps)。 Python 实现中的原始转写文件(`transcript-*.md`)则采用如下格式,时间戳由分段 `start`/`end` 计算为 `MM:SS`: ```markdown # Transcrição de Áudio **Arquivo:** meeting.mp3 **Idioma:** PT **Data:** 2026-02-03 02:30:45 --- ## Transcrição Completa **[00:12 → 00:45]** Good morning everyone. Let's start with updates from the frontend team. **[00:46 → 01:23]** We completed the dashboard redesign and deployed to staging yesterday.文件名与防覆盖机制v1.1.0 起改用时间戳命名,彻底解决重复运行覆盖问题(CHANGELOG 中记录为修复项 #5):TIMESTAMP=$(date +%Y%m%d-%H%M%S) TRANSCRIPT_FILE="transcript-${TIMESTAMP}.md" ATA_FILE="ata-${TIMESTAMP}.md"Python 侧save_outputs()用datetime.now().strftime("%Y%m%d-%H%M%S")生成时间戳,始终先保存transcript-*.md,若 LLM 生成了纪要则额外保存ata-*.md,且文件写入均通过safe_user_path()校验路径不越出当前工作区(相对路径、expanduser展开后必须以Path.cwd()为前缀,否则抛出ValueError),从源码层面杜绝了路径穿越风险。LLM 智能摘要工作流这是 v1.1.0 的核心升级。技能通过detect_cli_tool()按优先级探测 LLM CLI:claude存在则优先,其次检测gh copilot是否可用,两者皆无则退化为纯转写模式(仅输出 transcript,并在结尾提示安装方式)。Scenario A:用户提供自定义提示词流程(详细指南 Step 3 与 transcribe.py 的handle_prompt_workflow()对应):以 Panel 展示用户原始提示词;若检测到~/.copilot/skills/prompt-engineer/SKILL.md存在,自动调用 prompt-engineer 优化提示词(不询问,直接改进);并排展示"优化版"与"原版"两份提示词;询问💡 Usar versão melhorada? [s/n],默认s;按所选版本继续处理。若 prompt-engineer 不可用,则直接使用用户原始提示词。Scenario B:无提示词时自动生成询问"Gerar prompt automaticamente?"(默认s,选n则跳过 LLM、仅输出 transcript);构造 meta-prompt 分析转写文本(取前 4000 字符),要求用 2-3 行建议:内容类型(会议/讲座/访谈)、推荐输出格式(正式纪要/执行摘要/结构化笔记)、推荐框架(RISEN/RODES/STAR 等);展示建议并确认格式;再让 prompt-engineer 基于建议生成完整结构化提示词;预览生成的提示词并最终确认;任何一步被拒绝都回退到内置的DEFAULT_MEETING_PROMPT。内置回退提示词(DEFAULT_MEETING_PROMPT)是一个完整的 RISEN 风格模板,指定了角色(专业文档化转写员)、指令(识别内容类型、提取主题、识别说话人、提取决策与行动项、组织 Markdown)、步骤与收窄规则(保持客观清晰、保留重要上下文、使用合适的 Markdown、必要时包含时间戳),保证无任何外部依赖时也能产出高质量纪要。LLM 调用与超时处理process_with_llm()把prompt + "\n\n---\n\nTranscrição:\n\n" + transcript拼为完整输入:Claude 路径:subprocess.run(['claude', '-'], input=full_prompt, ...),即通过 stdin 管道喂给 Claude CLI;Copilot 路径:['gh', 'copilot', 'suggest', '-t', 'shell', full_prompt];两者均设 300 秒(5 分钟)超时,失败或超时时打印错误摘要并返回None,主流程随即降级为仅保存 transcript,保证任务不中断。转写与 LLM 处理全程有可视化反馈:转写分段用tqdm进度条,LLM 处理用rich的SpinnerColumn转圈动画,状态消息按颜色区分(绿=成功、黄=警告、红=错误)。实际使用:触发短语与三种典型场景SKILL.md 与 README 共同定义了激活短语:"transcribe audio to markdown"、"transcreva este áudio"、"convert audio file to text"、"extract speech from audio"、"áudio para texto com metadados"(技能原生支持葡萄牙语指令,与示例中的 pt-BR 语言检测相呼应)。场景一:基础转写copilot> transcribe audio to markdown: meeting-2026-02-02.mp3SKILL.md 展示的完整输出流为:引擎与 ffmpeg 检测 → 文件元数据(大小 12.3 MB、时长 00:45:32)→ 处理进度条 → 语言检测(pt-BR)→ 识别 4 位说话人 → 生成 Markdown,最终产出meeting-2026-02-02.md,报告 6842 词、耗时 127s,并给出下一步建议(审阅纪要、分享报告、跟踪行动项)。场景二:批处理copilot> transcreva estes áudios: recordings/*.mp3SKILL.md 的批处理输出显示:先枚举目录内 5 个文件(team-standup、client-call、brainstorm-session、product-demo、retrospective),逐个转写并报告各自耗时,最终汇总"Generated 5 Markdown reports / Total processing time"。批处理同样支持限定类型,如transcribe audio: recordings/*.wav只处理 WAV 文件。场景三:大文件预警与确认copilot> transcribe audio to markdown: conference-keynote.mp3【免费下载链接】agentic-awesome-skillsAAS Core is the local, agent-first control plane for complete catalog discovery, agent-owned selection, stack validation, and planning, backed by 2,600+ agentic skills. Includes CLI, local MCP, catalog, plugins, and Workbench.项目地址: https://gitcode.com/GitHub_Trending/an/agentic-awesome-skills创作声明:本文部分内容由AI辅助生成(AIGC),仅供参考