agentsclimarketplace

Voice transcribe

Skill vdk888/bubble-skills/voice-transcribe

The Bubble Shop — free, open-source Claude Code skills & plugins by Bubble Invest. One-command install. We're not developers — we built these to run our own family office. Help us improve them.

Install
npx -y skills add vdk888/bubble-skills --skill voice-transcribe

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

One thing to look at

  • 0 stars0 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 a voice note or audio file to text locally (offline, no API/key) — Ogg/Opus (Telegram voice), WAV, MP3, FLAC. Use when a message arrives with a voice attachment, or whenever you need speech-to-text for an audio file. Outputs the transcript text; supports French/English/auto language detection.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

1.9 KB, 414 tokens by cl100k_base, as published. Nobody here has run it

Voice transcription (local, offline)

Transcribes audio to text on-device using whisper.cpp (small multilingual model) plus a pure-WASM Opus decoder. No network, no API key, works headless. Good for French and English.

Setup (one time)

bash ${CLAUDE_SKILL_DIR}/setup.sh

This installs the Node Opus decoder and builds whisper.cpp + downloads the small model (~466 MB) into the skill's engine/ folder. Idempotent — safe to re-run. Requires git, cmake, a C compiler, and node (18+).

When to use

  • A message arrives with a voice attachment (audio/ogg, Telegram .oga).
  • You have any audio file (.oga, .ogg, .wav, .mp3, .flac) and need its text.

How to use

${CLAUDE_SKILL_DIR}/bin/transcribe.sh <audio-path> [lang]
  • lang is optional: fr, en, or auto (default). Force fr/en for best accuracy.
  • The transcript is printed to stdout.

Example:

${CLAUDE_SKILL_DIR}/bin/transcribe.sh ~/Downloads/voice-note.oga fr

Notes

  • First word of output may have a leading space — trim if needed.
  • A ~15 s clip transcribes in a few seconds on CPU.
  • To point at an existing whisper.cpp build instead of the bundled one, export VT_ENGINE=/path/to/whisper.cpp (must contain build/bin/whisper-cli + models/ggml-small.bin).
  • For higher accuracy at the cost of speed/size, swap the model in setup.sh (e.g. medium).

What ships with it: 5 files

5.8 KB alongside SKILL.md, 3 of them executable

bin/

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.