Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
RAM-safe wrapper around OpenAI Whisper that automatically chunks audio files longer than 10 minutes to stay within 1.5GB RAM. Uses ffmpeg for splitting and merging, cleans up temp files on exit. Designed for 16GB machines running multiple services concurrently.
Unlike standard Whisper invocations that load the full model and crash 16GB systems, this skill caps RAM at 1.5GB by defaulting to the base model and splitting long audio into 10-minute chunks.
When to use it
- Transcribing Telegram voice messages without stopping other agents
- Converting a 45-minute interview recording to text on a laptop
- Processing WhatsApp voice notes in bulk overnight
- Transcribing podcast clips while Ollama runs in the background
- Getting quick text from voice memos before a meeting
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: transcribe this 30-minute interview recording at /recordings/interview.ogg
- 1Run ffprobe to detect audio length (30 min, exceeds 10-min threshold)
- 2Split audio into three 10-minute chunks via ffmpeg
- 3Transcribe each chunk sequentially using Whisper base model (~1.5GB RAM peak)
- 4Concatenate the three transcripts in order
- 5Clean up temp chunk files and return the merged transcript path
Full transcript text file saved to output directory with no gaps between chunks
Requirements
Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.