agentleFS
rumotionrumotion/ai-agent-project-template/.agents/skills/local-media-transcription/SKILL.md

local-media-transcription

Skill1 starsChanged 8 days agoMITView on GitHub

Transcribe local audio and video files using whisper large-v3 on the local NVIDIA RTX A5000 GPU with automatic fallback ladder (cuda/float16 -> cuda/int8_float16 -> cpu chunked). Outputs timestamped .txt and .srt files locally with zero cloud leakage.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill whenever you need to transcribe audio or video recordings (e.g.

    line 8
  2. Very specific:

    GPU-First (NVIDIA RTX A5000): Always execute transcription on CUDA via faster-whisper (CTranslate2).

    line 14
  3. Very specific:

    Model Constraint: Always use large-v3 (32 layers).

    line 15
  4. Distinctive:

    100% Local & Confidential: No audio, video, or transcripts may be sent to any external service or API.

    line 16
  5. Distinctive:

    Run the bundled transcription script on your target media file:

    line 26

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Local Media Transcription (GPU-First)

Use this skill whenever you need to transcribe audio or video recordings (e.g. meeting recordings, interviews, technical presentations, internal calls) in any project on this machine.


Standing Rules & Constraints

  1. GPU-First (NVIDIA RTX A5000): Always execute transcription on CUDA via faster-whisper (CTranslate2).
  2. Model Constraint: Always use large-v3 (32 layers). Never substitute turbo, small, base, or medium. Proper-noun fidelity (vendor names, model names, people) is mandatory.
  3. 100% Local & Confidential: No audio, video, or transcripts may be sent to any external service or API.
  4. Precision Fallback Ladder:
    • Tier 1: CUDA float16 (Default, ~4.0 GB VRAM, ~5x real-time speedup)
    • Tier 2: CUDA int8_float16 (Fallback on VRAM pressure, ~2.8 GB VRAM)
    • Tier 3: CPU chunked parallel (Emergency fallback if CUDA is unavailable)

How to Transcribe

Run the bundled transcription script on your target media file:

python .agents/skills/local-media-transcription/scripts/transcribe.py "<media-file>" [output-dir]

Or pass the direct path:

python scripts/transcribe_local_large_v3.py "<media-file>" [output-dir]

Outputs

  • <output-dir>/<stem>_large-v3.txt (Timestamped format: [HH:MM:SS] text)
  • <output-dir>/<stem>_large-v3.srt (Standard SRT subtitle format with millisecond precision)

Technical Details

  • Language Detection: Detects language once on the first 30 seconds using tiny and forces the detected language on large-v3 to prevent per-segment mis-detection.
  • Audio Extraction: Uses ffmpeg -vn -ac 1 -ar 16000 -c:a pcm_s16le for clean 16kHz mono audio.
  • Speech Conditioning: Sets condition_on_previous_text=False and vad_filter=False to prevent clipping quiet words or hallucinating across silence boundaries.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.