Voice Transcription

Voice Transcription Skill

Transcribe audio files (voice memos, recordings, podcasts) to text using OpenAI's Whisper API or local transcription tools.

Purpose

This skill enables PCP to:

• Transcribe voice memos sent via Discord
• Convert audio recordings to searchable text
• Process podcast clips or meeting recordings
• Handle any audio-to-text conversion

When to Use

• User sends a voice memo or audio file
• User asks to "transcribe this"
• Any audio file needs to be converted to text
• Voice notes need to be captured in the vault

Supported Formats

OpenAI Whisper supports: mp3, mp4, mpeg, mpga, m4a, wav, webm, ogg, flac

For other formats (like opus), ffmpeg will convert automatically.

How to Execute

Step 1: Identify the Audio File

Check if the file exists and identify its format:

file /path/to/audio.file

Step 2: Convert if Necessary

If the format isn't directly supported, convert with ffmpeg:

# Convert to WAV (most compatible)
ffmpeg -i input.opus -ar 16000 -ac 1 -y output.wav

# Common conversions:
ffmpeg -i input.ogg -y output.mp3
ffmpeg -i input.webm -y output.mp3

Step 3: Transcribe Using Helper Script

Use the bundled transcribe.py script:

python /workspace/skills/voice-transcription/transcribe.py /path/to/audio.wav

Or in Python:

from pathlib import Path
import sys
sys.path.insert(0, "/workspace/skills/voice-transcription")
from transcribe import transcribe_audio

result = transcribe_audio("/path/to/audio.wav")
if result["success"]:
    print(result["text"])
else:
    print(f"Error: {result['error']}")

Step 4: Return Results

Present the transcription to the user in a readable format. Optionally capture it to the vault:

from vault_v2 import smart_capture

# Capture the transcription
result = smart_capture(
    f"Voice memo transcription: {transcription_text}",
    capture_type="note"
)

Fallback: Local Transcription

If OPENAI_API_KEY is not available but local whisper CLI is installed:

# Check if whisper is installed
which whisper

# Use local whisper
whisper /path/to/audio.wav --model base --output_format txt

Configuration

Configure in /workspace/config/pcp.yaml:

skills:
  entries:
    voice-transcription:
      enabled: true
      model: whisper-1           # OpenAI model
      language: en               # Default language (optional, auto-detect if not set)
      fallback_to_local: true    # Use local whisper if API fails
      max_file_size_mb: 25       # Maximum file size to process

Error Handling

Example Workflow

User sends voice memo via Discord:

1. Discord attachment saved to /tmp/discord_attachments/voice-memo.ogg
2. PCP detects audio file, activates voice-transcription skill
3. Convert to supported format: ffmpeg -i voice-memo.ogg -y voice-memo.mp3
4. Transcribe: transcribe_audio("voice-memo.mp3")
5. Return text to user: "Here's what you said: ..."
6. Optionally capture to vault

Related Skills

• /vault-operations - Capture transcriptions
• /email-processing - Transcribe voice messages from emails
• /task-delegation - Delegate long audio files to background processing