Audio Transcriber Pro

\r \r

Purpose\r

\r This skill automates audio-to-text transcription with professional Markdown output, extracting rich technical metadata (speakers, timestamps, language, file size, duration) and generating structured meeting minutes and executive summaries. It uses Faster-Whisper or Whisper with zero configuration, working universally across projects without hardcoded paths or API keys.\r \r Inspired by tools like Plaud, this skill transforms raw audio recordings into actionable documentation, making it ideal for meetings, interviews, lectures, and content analysis.\r \r

When to Use\r

\r Invoke this skill when:\r \r

• User needs to transcribe audio/video files to text\r
• User wants meeting minutes automatically generated from recordings\r
• User requires speaker identification (diarization) in conversations\r
• User needs subtitles/captions (SRT, VTT formats)\r
• User wants executive summaries of long audio content\r
• User asks variations of "transcribe this audio", "convert audio to text", "generate meeting notes from recording"\r
• User has audio files in common formats (MP3, WAV, M4A, OGG, FLAC, WEBM)\r \r

Workflow\r

\r

Step 0: Discovery (Auto-detect Transcription Tools)\r

\r Objective: Identify available transcription engines without user configuration.\r \r Actions:\r \r Run detection commands to find installed tools:\r \r

# Check for Faster-Whisper (preferred - 4-5x faster)\r
if python3 -c "import faster_whisper" 2>/dev/null; then\r
    TRANSCRIBER="faster-whisper"\r
    echo "✅ Faster-Whisper detected (optimized)"\r
# Fallback to original Whisper\r
elif python3 -c "import whisper" 2>/dev/null; then\r
    TRANSCRIBER="whisper"\r
    echo "✅ OpenAI Whisper detected"\r
else\r
    TRANSCRIBER="none"\r
    echo "⚠️  No transcription tool found"\r
fi\r
\r
# Check for ffmpeg (audio format conversion)\r
if command -v ffmpeg &>/dev/null; then\r
    echo "✅ ffmpeg available (format conversion enabled)"\r
else\r
    echo "ℹ️  ffmpeg not found (limited format support)"\r
fi\r
```\r
\r
**If no transcriber found:**\r
\r
Offer automatic installation using the provided script:\r
\r
```bash\r
echo "⚠️  No transcription tool found"\r
echo ""\r
echo "🔧 Auto-install dependencies? (Recommended)"\r
read -p "Run installation script? [Y/n]: " AUTO_INSTALL\r
\r
if [[ ! "$AUTO_INSTALL" =~ ^[Nn] ]]; then\r
    # Get skill directory (works for both repo and symlinked installations)\r
    SKILL_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"\r
    \r
    # Run installation script\r
    if [[ -f "$SKILL_DIR/scripts/install-requirements.sh" ]]; then\r
        bash "$SKILL_DIR/scripts/install-requirements.sh"\r
    else\r
        echo "❌ Installation script not found"\r
        echo ""\r
        echo "📦 Manual installation:"\r
        echo "  pip install faster-whisper  # Recommended"\r
        echo "  pip install openai-whisper  # Alternative"\r
        echo "  brew install ffmpeg         # Optional (macOS)"\r
        exit 1\r
    fi\r
    \r
    # Verify installation succeeded\r
    if python3 -c "import faster_whisper" 2>/dev/null || python3 -c "import whisper" 2>/dev/null; then\r
        echo "✅ Installation successful! Proceeding with transcription..."\r
    else\r
        echo "❌ Installation failed. Please install manually."\r
        exit 1\r
    fi\r
else\r
    echo ""\r
    echo "📦 Manual installation required:"\r
    echo ""\r
    echo "Recommended (fastest):"\r
    echo "  pip install faster-whisper"\r
    echo ""\r
    echo "Alternative (original):"\r
    echo "  pip install openai-whisper"\r
    echo ""\r
    echo "Optional (format conversion):"\r
    echo "  brew install ffmpeg  # macOS"\r
    echo "  apt install ffmpeg   # Linux"\r
    echo ""\r
    exit 1\r
fi\r
```\r
\r
This ensures users can install dependencies with one confirmation, or opt for manual installation if preferred.\r
\r
**If transcriber found:**\r
\r
Proceed to Step 0b (CLI Detection).\r
\r
\r
### Step 1: Validate Audio File\r
\r
**Objective:** Verify file exists, check format, and extract metadata.\r
\r
**Actions:**\r
\r
1. **Accept file path or URL** from user:\r
   - Local file: `meeting.mp3`\r
   - URL: `https://example.com/audio.mp3` (download to temp directory)\r
\r
2. **Verify file exists:**\r
\r
```bash\r
if [[ ! -f "$AUDIO_FILE" ]]; then\r
    echo "❌ File not found: $AUDIO_FILE"\r
    exit 1\r
fi\r
```\r
\r
3. **Extract metadata** using ffprobe or file utilities:\r
\r
```bash\r
# Get file size\r
FILE_SIZE=$(du -h "$AUDIO_FILE" | cut -f1)\r
\r
# Get duration and format using ffprobe\r
DURATION=$(ffprobe -v error -show_entries format=duration \\r
    -of default=noprint_wrappers=1:nokey=1 "$AUDIO_FILE" 2>/dev/null)\r
FORMAT=$(ffprobe -v error -select_streams a:0 -show_entries \\r
    stream=codec_name -of default=noprint_wrappers=1:nokey=1 "$AUDIO_FILE" 2>/dev/null)\r
\r
# Convert duration to HH:MM:SS\r
DURATION_HMS=$(date -u -r "$DURATION" +%H:%M:%S 2>/dev/null || echo "Unknown")\r
```\r
\r
4. **Check file size** (warn if large for cloud APIs):\r
\r
```bash\r
SIZE_MB=$(du -m "$AUDIO_FILE" | cut -f1)\r
if [[ $SIZE_MB -gt 25 ]]; then\r
    echo "⚠️  Large file ($FILE_SIZE) - processing may take several minutes"\r
fi\r
```\r
\r
5. **Validate format** (supported: MP3, WAV, M4A, OGG, FLAC, WEBM):\r
\r
```bash\r
EXTENSION="${AUDIO_FILE##*.}"\r
SUPPORTED_FORMATS=("mp3" "wav" "m4a" "ogg" "flac" "webm" "mp4")\r
\r
if [[ ! " ${SUPPORTED_FORMATS[@]} " =~ " ${EXTENSION,,} " ]]; then\r
    echo "⚠️  Unsupported format: $EXTENSION"\r
    if command -v ffmpeg &>/dev/null; then\r
        echo "🔄 Converting to WAV..."\r
        ffmpeg -i "$AUDIO_FILE" -ar 16000 "${AUDIO_FILE%.*}.wav" -y\r
        AUDIO_FILE="${AUDIO_FILE%.*}.wav"\r
    else\r
        echo "❌ Install ffmpeg to convert formats: brew install ffmpeg"\r
        exit 1\r
    fi\r
fi\r
```\r
\r
\r
### Step 3: Generate Markdown Output\r
\r
**Objective:** Create structured Markdown with metadata, transcription, meeting minutes, and summary.\r
\r
**Output Template:**\r
\r
```markdown\r
# Audio Transcription Report\r
\r
## 📊 Metadata\r
\r
| Field | Value |\r
|-------|-------|\r
| **File Name** | {filename} |\r
| **File Size** | {file_size} |\r
| **Duration** | {duration_hms} |\r
| **Language** | {language} ({language_code}) |\r
| **Processed Date** | {process_date} |\r
| **Speakers Identified** | {num_speakers} |\r
| **Transcription Engine** | {engine} (model: {model}) |\r
\r
\r
## 📋 Meeting Minutes\r
\r
### Participants\r
- {speaker_1}\r
- {speaker_2}\r
- ...\r
\r
### Topics Discussed\r
1. **{topic_1}** ({timestamp})\r
   - {key_point_1}\r
   - {key_point_2}\r
\r
2. **{topic_2}** ({timestamp})\r
   - {key_point_1}\r
\r
### Decisions Made\r
- ✅ {decision_1}\r
- ✅ {decision_2}\r
\r
### Action Items\r
- [ ] **{action_1}** - Assigned to: {speaker} - Due: {date_if_mentioned}\r
- [ ] **{action_2}** - Assigned to: {speaker}\r
\r
\r
*Generated by audio-transcriber skill v1.0.0*  \r
*Transcription engine: {engine} | Processing time: {elapsed_time}s*\r
```\r
\r
**Implementation:**\r
\r
Use Python or bash with AI model (Claude/GPT) for intelligent summarization:\r
\r
```python\r
def generate_meeting_minutes(segments):\r
    """Extract topics, decisions, action items from transcription."""\r
    \r
    # Group segments by topic (simple clustering by timestamps)\r
    topics = cluster_by_topic(segments)\r
    \r
    # Identify action items (keywords: "should", "will", "need to", "action")\r
    action_items = extract_action_items(segments)\r
    \r
    # Identify decisions (keywords: "decided", "agreed", "approved")\r
    decisions = extract_decisions(segments)\r
    \r
    return {\r
        "topics": topics,\r
        "decisions": decisions,\r
        "action_items": action_items\r
    }\r
\r
def generate_summary(segments, max_paragraphs=5):\r
    """Create executive summary using AI (Claude/GPT via API or local model)."""\r
    \r
    full_text = " ".join([s["text"] for s in segments])\r
    \r
    # Use Chain of Density approach (from prompt-engineer frameworks)\r
    summary_prompt = f"""\r
    Summarize the following transcription in {max_paragraphs} concise paragraphs.\r
    Focus on key topics, decisions, and action items.\r
    \r
    Transcription:\r
    {full_text}\r
    """\r
    \r
    # Call AI model (placeholder - user can integrate Claude API or use local model)\r
    summary = call_ai_model(summary_prompt)\r
    \r
    return summary\r
```\r
\r
**Output file naming:**\r
\r
```bash\r
# v1.1.0: Use timestamp para evitar sobrescrever\r
TIMESTAMP=$(date +%Y%m%d-%H%M%S)\r
TRANSCRIPT_FILE="transcript-${TIMESTAMP}.md"\r
ATA_FILE="ata-${TIMESTAMP}.md"\r
\r
echo "$TRANSCRIPT_CONTENT" > "$TRANSCRIPT_FILE"\r
echo "✅ Transcript salvo: $TRANSCRIPT_FILE"\r
\r
if [[ -n "$ATA_CONTENT" ]]; then\r
    echo "$ATA_CONTENT" > "$ATA_FILE"\r
    echo "✅ Ata salva: $ATA_FILE"\r
fi\r
```\r
\r
\r
#### **SCENARIO A: User Provided Custom Prompt**\r
\r
**Workflow:**\r
\r
1. **Display user's prompt:**\r
   ```\r
   📝 Prompt fornecido pelo usuário:\r
   ┌──────────────────────────────────┐\r
   │ [User's prompt preview]          │\r
   └──────────────────────────────────┘\r
   ```\r
\r
2. **Automatically improve with prompt-engineer (if available):**\r
   ```bash\r
   🔧 Melhorando prompt com prompt-engineer...\r
   [Invokes: gh copilot -p "melhore este prompt: {user_prompt}"]\r
   ```\r
\r
3. **Show both versions:**\r
   ```\r
   ✨ Versão melhorada:\r
   ┌──────────────────────────────────┐\r
   │ Role: Você é um documentador...  │\r
   │ Instructions: Transforme...      │\r
   │ Steps: 1) ... 2) ...             │\r
   │ End Goal: ...                    │\r
   └──────────────────────────────────┘\r
\r
   📝 Versão original:\r
   ┌──────────────────────────────────┐\r
   │ [User's original prompt]         │\r
   └──────────────────────────────────┘\r
   ```\r
\r
4. **Ask which to use:**\r
   ```bash\r
   💡 Usar versão melhorada? [s/n] (default: s):\r
   ```\r
\r
5. **Process with selected prompt:**\r
   - If "s": use improved\r
   - If "n": use original\r
\r
\r
#### **LLM Processing (Both Scenarios)**\r
\r
Once prompt is finalized:\r
\r
```python\r
from rich.progress import Progress, SpinnerColumn, TextColumn\r
\r
def process_with_llm(transcript, prompt, cli_tool='claude'):\r
    full_prompt = f"{prompt}\
\
---\
\
Transcrição:\
\
{transcript}"\r
    \r
    with Progress(\r
        SpinnerColumn(),\r
        TextColumn("[progress.description]{task.description}"),\r
        transient=True\r
    ) as progress:\r
        progress.add_task(\r
            description=f"🤖 Processando com {cli_tool}...",\r
            total=None\r
        )\r
        \r
        if cli_tool == 'claude':\r
            result = subprocess.run(\r
                ['claude', '-'],\r
                input=full_prompt,\r
                capture_output=True,\r
                text=True,\r
                timeout=300  # 5 minutes\r
            )\r
        elif cli_tool == 'gh-copilot':\r
            result = subprocess.run(\r
                ['gh', 'copilot', 'suggest', '-t', 'shell', full_prompt],\r
                capture_output=True,\r
                text=True,\r
                timeout=300\r
            )\r
    \r
    if result.returncode == 0:\r
        return result.stdout.strip()\r
    else:\r
        return None\r
```\r
\r
**Progress output:**\r
```\r
🤖 Processando com claude... ⠋\r
[After completion:]\r
✅ Ata gerada com sucesso!\r
```\r
\r
\r
#### **Final Output**\r
\r
**Success (both files):**\r
```bash\r
💾 Salvando arquivos...\r
\r
✅ Arquivos criados:\r
  - transcript-20260203-023045.md  (transcript puro)\r
  - ata-20260203-023045.md         (processado com LLM)\r
\r
🧹 Removidos arquivos temporários: metadata.json, transcription.json\r
\r
✅ Concluído! Tempo total: 3m 45s\r
```\r
\r
**Transcript only (user declined LLM):**\r
```bash\r
💾 Salvando arquivos...\r
\r
✅ Arquivo criado:\r
  - transcript-20260203-023045.md\r
\r
ℹ️  Ata não gerada (processamento LLM recusado pelo usuário)\r
\r
🧹 Removidos arquivos temporários: metadata.json, transcription.json\r
\r
✅ Concluído!\r
```\r
\r
\r
### Step 5: Display Results Summary\r
\r
**Objective:** Show completion status and next steps.\r
\r
**Output:**\r
\r
```bash\r
echo ""\r
echo "✅ Transcription Complete!"\r
echo ""\r
echo "📊 Results:"\r
echo "  File: $OUTPUT_FILE"\r
echo "  Language: $LANGUAGE"\r
echo "  Duration: $DURATION_HMS"\r
echo "  Speakers: $NUM_SPEAKERS"\r
echo "  Words: $WORD_COUNT"\r
echo "  Processing time: ${ELAPSED_TIME}s"\r
echo ""\r
echo "📝 Generated:"\r
echo "  - $OUTPUT_FILE (Markdown report)"\r
[if alternative formats:]\r
echo "  - ${OUTPUT_FILE%.*}.srt (Subtitles)"\r
echo "  - ${OUTPUT_FILE%.*}.json (Structured data)"\r
echo ""\r
echo "🎯 Next steps:"\r
echo "  1. Review meeting minutes and action items"\r
echo "  2. Share report with participants"\r
echo "  3. Track action items to completion"\r
```\r
\r
\r
## Example Usage\r
\r
### **Example 1: Basic Transcription**\r
\r
**User Input:**\r
```bash\r
copilot> transcribe audio to markdown: meeting-2026-02-02.mp3\r
```\r
\r
**Skill Output:**\r
\r
```bash\r
✅ Faster-Whisper detected (optimized)\r
✅ ffmpeg available (format conversion enabled)\r
\r
📂 File: meeting-2026-02-02.mp3\r
📊 Size: 12.3 MB\r
⏱️  Duration: 00:45:32\r
\r
🎙️  Processing...\r
[████████████████████] 100%\r
\r
✅ Language detected: Portuguese (pt-BR)\r
👥 Speakers identified: 4\r
📝 Generating Markdown output...\r
\r
✅ Transcription Complete!\r
\r
📊 Results:\r
  File: meeting-2026-02-02.md\r
  Language: pt-BR\r
  Duration: 00:45:32\r
  Speakers: 4\r
  Words: 6,842\r
  Processing time: 127s\r
\r
📝 Generated:\r
  - meeting-2026-02-02.md (Markdown report)\r
\r
🎯 Next steps:\r
  1. Review meeting minutes and action items\r
  2. Share report with participants\r
  3. Track action items to completion\r
```\r
\r
\r
### **Example 3: Batch Processing**\r
\r
**User Input:**\r
```bash\r
copilot> transcreva estes áudios: recordings/*.mp3\r
```\r
\r
**Skill Output:**\r
\r
```bash\r
📦 Batch mode: 5 files found\r
  1. team-standup.mp3\r
  2. client-call.mp3\r
  3. brainstorm-session.mp3\r
  4. product-demo.mp3\r
  5. retrospective.mp3\r
\r
🎙️  Processing batch...\r
\r
[1/5] team-standup.mp3 ✅ (2m 34s)\r
[2/5] client-call.mp3 ✅ (15m 12s)\r
[3/5] brainstorm-session.mp3 ✅ (8m 47s)\r
[4/5] product-demo.mp3 ✅ (22m 03s)\r
[5/5] retrospective.mp3 ✅ (11m 28s)\r
\r
✅ Batch Complete!\r
📝 Generated 5 Markdown reports\r
⏱️  Total processing time: 6m 15s\r
```\r
\r
\r
### **Example 5: Large File Warning**\r
\r
**User Input:**\r
```bash\r
copilot> transcribe audio to markdown: conference-keynote.mp3\r
```\r
\r
**Skill Output:**\r
\r
```bash\r
✅ Faster-Whisper detected (optimized)\r
\r
📂 File: conference-keynote.mp3\r
📊 Size: 87.2 MB\r
⏱️  Duration: 02:15:47\r
⚠️  Large file (87.2 MB) - processing may take several minutes\r
\r
Continue? [Y/n]:\r
```\r
\r
**User:** `Y`\r
\r
```bash\r
🎙️  Processing... (this may take 10-15 minutes)\r
[████░░░░░░░░░░░░░░░░] 20% - Estimated time remaining: 12m\r
```\r
\r
\r
This skill is **platform-agnostic** and works in any terminal context where GitHub Copilot CLI is available. It does not depend on specific project configurations or external APIs, following the zero-configuration philosophy.\r