Voice transcribe

Skill jehrhardt/agents/skills/voice-transcribe

My agent skills

Install

npx -y skills add jehrhardt/agents --skill voice-transcribe

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Transcribe AAC/m4a/audio files to text using faster-whisper. Use when the user wants to convert speech recordings into text.

SKILL.md

2.2 KB, 515 tokens by cl100k_base, as published. Nobody here has run it

Voice Transcribe

Convert audio files to text using faster-whisper (Python), managed entirely by uv.

Prerequisites

uv installed (used to run the script — it auto-creates the environment and installs dependencies)
ffmpeg installed (for converting non-native audio formats like AAC/M4A)
The script has inline PEP 723 metadata declaring its Python version (≥3.14) and dependencies, so no manual setup is needed

Steps

1. Verify the Input File

file "<path>"

2. Transcribe

Run the script via uv. It will automatically download Python 3.14 (if missing), create a virtual environment, install dependencies, and run transcription:

uv run ~/.agents/skills/voice-transcribe/transcribe.py "<path>"

With language and device options:

uv run ~/.agents/skills/voice-transcribe/transcribe.py "<path>" --language de --device cuda --compute-type float16

With timestamps:

uv run ~/.agents/skills/voice-transcribe/transcribe.py "<path>" --timestamps

For all options, run:

uv run ~/.agents/skills/voice-transcribe/transcribe.py --help

3. Review

This step is mandatory. Present the full transcript to the user and call out likely errors:

Gibberish words or obvious hallucinations
English loanwords that look misheard (e.g., "Auto Mash" for "Automerge")
German compound words that seem broken
Technical terms that appear garbled
Sentences that don't make grammatical sense

Wait for the user's response. If they want fixes, apply them. If they want to review themselves, present the raw transcript and proceed only after explicit confirmation.

4. Use the Transcription

Only after the user has confirmed or approved the transcript, use it for the requested purpose (append to daily notes, summarize, create wiki pages, etc.).

Notes

Supported inputs: any audio/video format ffmpeg can decode (mp3, wav, m4a, aac, ogg, flac, mp4, etc.)
Non-native formats (AAC, M4A, MP4, MOV, etc.) are automatically converted to WAV using ffmpeg
Models are downloaded automatically on first use (cached in HuggingFace cache)