When to Use This Skill

/voice-meeting

Process voice meeting recordings or voice-to-text transcripts into structured meeting notes. Applies speech-to-text corrections for common misrecognitions and generates structured output with decisions and actions.

When to Use This Skill

• Processing transcripts from voice recording apps
• Cleaning up speech-to-text output from meeting recordings
• Converting voice notes into structured meeting records
• Processing dictated meeting summaries

Usage

/voice-meeting "<meeting-title>" [--source transcript|voicenote|dictation] [--corrections path/to/corrections.md]

Parameters

Instructions

Phase 1: Ingest Transcript

1. Accept the raw transcript from the user — this may come from:
   • Voice recording app transcript
   • Dictation software output
   • Manual transcription with errors
2. Load correction rules — If --corrections provided, load custom corrections. Otherwise, apply standard speech-to-text corrections:
   • Common homophones (their/there, your/you're)
   • Proper noun corrections (specific to the user's domain)
   • Technical term corrections
3. Ask for context:
   • Date and time of the meeting
   • Attendees (the transcript may not identify speakers)
   • Related project or topic

Phase 2: Clean and Correct

Apply corrections systematically:

1. Proper nouns — Fix misrecognised names, system names, and project names
2. Technical terms — Fix misrecognised technical vocabulary
3. Sentence boundaries — Fix run-on sentences and paragraph breaks
4. Speaker attribution — If speakers are identified, format as dialogue
5. Filler removal — Remove "um", "uh", "like", "you know" etc.
6. Grammar — Fix obvious grammatical errors from speech patterns

Phase 3: Structure into Meeting Note

Follow the same structure as /meeting-notes:

1. Extract topics — Identify distinct discussion topics from the conversation flow
2. Extract decisions — Find decision language
3. Extract actions — Find action assignments
4. Generate summary — Summarise the overall meeting
5. Add metadata — Date, attendees, project, tags

Output Format

---
type: Meeting
title: "Meeting - YYYY-MM-DD <Meeting Title>"
created: YYYY-MM-DD
date: YYYY-MM-DD
attendees:
  - "[[Person Name]]"
source: voice-transcript
tags: [activity/meeting, project/relevant-tag]
summary: "<One-line summary>"
---

# Meeting - YYYY-MM-DD <Meeting Title>

> **Source:** Voice transcript | **Corrections applied:** X

## Attendees

- [[Person 1]]
- [[Person 2]]

## Discussion

### <Topic 1>

<Cleaned and structured discussion summary>

### <Topic 2>

<Cleaned and structured discussion summary>

## Decisions

| # | Decision            | Deciders     |
|---|---------------------|-------------|
| 1 | <Decision>          | <Who>       |

## Action Items

| # | Action              | Owner        | Deadline    |
|---|---------------------|-------------|-------------|
| 1 | <Action>            | [[Person]]  | YYYY-MM-DD  |

## Transcript Corrections Applied

| Original               | Corrected To           | Reason             |
|------------------------|------------------------|--------------------|
| <misrecognised text>   | <corrected text>       | Proper noun        |

Examples

Example 1: Voice App Transcript

/voice-meeting "Sprint Retrospective" --source transcript

User pastes transcript from a voice recording app; skill cleans and structures it.

Example 2: With Custom Corrections

/voice-meeting "Architecture Review" --corrections .claude/rules/transcript-corrections.md

Applies domain-specific corrections from a custom corrections file.

Example 3: Dictated Summary

/voice-meeting "Quick Sync" --source dictation

Processes a dictated meeting summary into structured notes.

Invoke with: /voice-meeting "<title>" then paste the raw transcript