Audio Transcribe

Auto-transcribe voice messages locally using faster-whisper with selectable Whisper models, no API key required.

Security Scan

The skill's code and instructions match its stated purpose (local transcription with faster-whisper); it does not request credentials or access unrelated resources.

Purpose & Capability

Name/description (local transcription with faster-whisper) align with included script and SKILL.md. The script only loads a WhisperModel and transcribes a provided audio file.

Instruction Scope

Runtime instructions and the script operate only on the provided audio path and model downloads; there are no references to unrelated files, environment variables, or external endpoints in the skill code itself.

Install Mechanism

No install spec in the registry; SKILL.md asks the user to pip install faster-whisper. This is expected but means a third-party package (and its dependencies) will be installed from PyPI and models will be downloaded from the network on first run.

Credentials

The skill requests no environment variables, credentials, or config paths. The script only uses the command-line audio file argument.

Persistence & Privilege

always is false and the skill does not modify agent configuration or request persistent privileges.

Assessment

This skill appears coherent for local transcription. Before installing: (1) be aware pip install faster-whisper will install third-party code from PyPI and may build native dependencies—review that package if you require strict supply-chain controls; (2) model files are downloaded on first use (network bandwidth, disk space, and cache location like ~/.cache may be used); (3) transcription can require significant RAM/CPU/GPU depending on model size—ensure your system meets the listed VRAM/RAM requirements; (4) the included script only reads the audio file you pass and prints the text (no hidden endpoints or credential access were found), but if you must guarantee full offline operation, pre-download models and verify faster-whisper’s sources before running.

Audio Transcription Skill

Auto-transcribe voice messages using faster-whisper (local, no API key needed).

Requirements

pip install faster-whisper

Models download automatically on first use.

Usage

Transcribe a file

python3 /root/clawd/skills/audio-transcribe/scripts/transcribe.py /path/to/audio.ogg

Change model (edit script)

Edit transcribe.py and change:

model = WhisperModel('small', device='cpu', compute_type='int8')  # Options: tiny, base, small, medium, large-v3

Models

Model	Size	VRAM/RAM	Speed	Use Case
tiny	39 MB	~1 GB	⚡⚡⚡	Quick drafts
base	74 MB	~1 GB	⚡⚡	Basic accuracy
small	244 MB	~2 GB	⚡	Recommended
medium	769 MB	~5 GB	🐢	Better accuracy
large-v3	1.5 GB	~10 GB	🐢🐢	Best accuracy

Integration

Clawdbot auto-transcribes incoming voice messages when this skill is enabled.

Files

• scripts/transcribe.py — Main transcription script
• SKILL.md — This file