Imported from lerboi/plagiacheck (.claude/skills/plagiacheck-tool-audio-summarizer/SKILL.md). Install upstream with npx skills add lerboi/plagiacheck --skill plagiacheck-tool-audio-summarizer. Copyright stays with the author.

Audio Summarizer

Records or accepts a pasted transcript, then produces a structured summary tailored to the content type (lecture, interview, meeting, podcast, speech).

Status: functional but browser-dependent — same Web Speech API caveats as the other voice tools.

Files

API contract

POST /api/voice-tools
{
  "text": "<transcript>",
  "tool": "audio-summarize"
}

Returns:

{
  "result": {
    "title": "...",
    "overview": "1-2 sentence overview",
    "keyPoints": ["point 1", "point 2", "point 3"],
    "detailedSummary": "...",
    "contentType": "lecture" | "interview" | "meeting" | "podcast" | "speech" | "other",
    "actionItems": ["..."]
  },
  "remainingTokens": <int>,
  "tokensUsed": <int>
}

Token cost

Math.ceil(text.length / 6) text tokens.

Prompt rules

The system prompt instructs the model to:

• Identify content type and tailor the summary (meetings get action items + decisions; interviews get Q&A pairs; lectures get key concepts/definitions).
• Aim for ~20-30% of the original length.
• Strip filler words and repetition.
• Preserve named entities, numbers, dates, and important quotes.

Browser caveats

Same as the other voice tools. Users can also paste a transcript directly without recording.

Common edits

• The page renders all four sections (overview, key points, action items, detailed summary). Hide unused sections (e.g. action items for podcasts) by checking result.contentType if you want a cleaner UI.