Skills audio-transcriber
📝

audio-transcriber

Safe ⚙️ External commands📁 Filesystem access🌐 Network access

Transcribe Audio to Markdown

Converts audio recordings into searchable text and generates professional meeting notes with AI-powered summaries using Claude or GitHub Copilot integration.

Supports: Claude Codex Code(CC)
🥈 81 Silver
1

Download the skill ZIP

2

Upload in Claude

Go to Settings → Capabilities → Skills → Upload skill

3

Toggle on and start using

Test it

Using "audio-transcriber". meeting-recording.mp3

Expected outcome:

# Transcrição de Áudio

**Arquivo:** meeting-recording.mp3
**Idioma:** en
**Data:** 2024-01-15

---

## Transcrição Completa

**[00:00 → 00:15]** Welcome everyone to our quarterly planning meeting.

**[00:15 → 00:45]** Today we will discuss the roadmap for Q2 and assign tasks...

---

## Ata Gerada

**Resumo:** This quarterly planning meeting covered Q2 roadmap items including product launch timeline, resource allocation, and key milestones.

**Decisões:**
- Approved Q2 launch date of April 15th
- Team Alpha assigned to feature development
- Weekly sync meetings scheduled for Tuesdays

Using "audio-transcriber". lecture-audio.wav

Expected outcome:

# Transcrição de Áudio

**Arquivo:** lecture-audio.wav
**Idioma:** pt
**Data:** 2024-01-20

---

## Pontos-Chave

1. **Conceitos Fundamentais** - Explanation of core concepts
2. **Aplicações Práticas** - Real-world applications
3. **Próximos Passos** - Recommended follow-up activities

Security Audit

Safe
v1 • 2/24/2026

Static analysis flagged 266 potential issues, but evaluation reveals these are primarily false positives. The flagged patterns in documentation files (README.md, SKILL.md, CHANGELOG.md) represent code examples, not executable code. The actual Python script (transcribe.py) uses hardcoded CLI tool invocations for legitimate transcription purposes. No command injection vectors or malicious intent detected. This is a standard audio transcription utility.

7
Files scanned
2,321
Lines analyzed
6
findings
1
Total audits
Medium Risk Issues (1)
External Command Execution in Documentation Examples
Static scanner flagged shell command examples in documentation files (README.md, SKILL.md, CHANGELOG.md). These are legitimate code examples for users to follow, not actual execution vectors.
Low Risk Issues (2)
Subprocess Calls with Hardcoded Arguments
The transcribe.py script uses subprocess.run to invoke CLI tools (claude, gh copilot). All arguments are hardcoded strings, not user-controlled input.
File System Operations for Output
Script writes transcription outputs to files. File paths are derived from input filename with timestamp appended.

Risk Factors

⚙️ External commands (2)
📁 Filesystem access (2)
🌐 Network access
No specific locations recorded
Audited by: claude

Quality Score

77
Architecture
100
Maintainability
87
Content
38
Community
96
Security
87
Spec Compliance

What You Can Build

Meeting Transcription

Automatically transcribe meeting recordings and generate professional meeting minutes with action items and decisions documented.

Podcast and Interview Notes

Convert podcast episodes or interview recordings into searchable text with AI-generated summaries for content creators.

Lecture and Course Transcription

Transform educational audio content into structured notes with key points extracted for study purposes.

Try These Prompts

Basic Audio Transcription
Transcribe this audio file to text. Output the complete transcription in Markdown format.
Meeting Minutes Generation
Analyze this meeting transcription and create structured meeting minutes including: attendees, agenda topics discussed, key decisions made, and action items with owners.
Executive Summary
Read this transcript and create a concise executive summary highlighting the main points, key takeaways, and any important announcements. Use bullet points for easy reading.
Q&A Documentation
Extract all questions and answers from this interview or Q&A session. Format as a structured document with questions clearly separated from answers.

Best Practices

  • Use high-quality audio recordings for better transcription accuracy - minimize background noise
  • Start with the base Whisper model for fastest processing, upgrade to small or medium for improved accuracy
  • Provide custom prompts to guide the LLM output toward your desired format (meeting notes, summaries, Q&A)

Avoid

  • Do not expect perfect transcription of poor-quality audio with significant background noise
  • Avoid processing very long files without chunking - may cause memory issues
  • Do not skip the model download step if you want offline capability

Frequently Asked Questions

What audio formats are supported?
The skill supports common audio formats including MP3, WAV, M4A, FLAC, and OGG. Any format readable by ffmpeg can be processed.
Do I need to install Whisper manually?
No, the install-requirements.sh script will automatically install faster-whisper or openai-whisper. The base model will be downloaded on first use.
Which is faster - faster-whisper or whisper?
Faster-Whisper is significantly faster and more memory-efficient as it uses CTranslate2. The script prioritizes faster-whisper and falls back to whisper if needed.
Can I use this offline?
Yes, after downloading the Whisper model, transcription works completely offline. AI summarization requires Claude CLI or GitHub Copilot CLI to be installed.
How accurate is the transcription?
Accuracy depends on audio quality and the model used. The base model provides good accuracy for clear audio. For better results, use small or medium models.
What if neither Claude nor Copilot is installed?
The script will detect missing CLIs and save only the transcription text without AI-generated summaries. You can add summaries later with Claude or Copilot.