agentleFS
godot-fungodot-fun/godot-agent/.agents/skills/audio-to-text/SKILL.md

audio-to-text

Skill210 starsChanged 2 days agoMITView on GitHub

Transcribes a single PCM16 WAV file to plain UTF-8 text locally through a standard-library Python wrapper around native SenseVoice Small F16 GGUF FunASR llama.cpp runtimes, preferring cross-vendor Vulkan GPU acceleration and automatically falling back to CPU, with built-in FSMN-VAD. Use when the user wants offline speech-to-text, audio transcription, ASR, or spoken-content extraction in Mandarin, Cantonese, English, Japanese, or Korean.

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 3 are distinctive, and 3 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.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read and follow dependency-manager before running commands.

    line 12
  2. Very specific:

    Run .ai/audio-to-text/transcribe.py through the default python manifest entry.

    line 14
  3. Very specific:

    Always pass --vad for long-audio segmentation.

    line 19
  4. Distinctive:

    Never overwrite or modify the source audio.

    line 21
  5. Distinctive:

    Do not write a transcript file.

    line 31
  6. Very specific:

    Use --cpu only when the user explicitly requests CPU or Vulkan troubleshooting requires bypassing GPU acceleration.

    line 33

≈ “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

Audio to Text

Run the bundled standard-library Python wrapper, which validates input, invokes native llama-funasr-sensevoice, removes runtime logs, and prints only UTF-8 transcription text. Prefer the Vulkan runtime for NVIDIA, AMD, or Intel GPUs and automatically retry with CPU when Vulkan is unavailable. Do not use CUDA, PyTorch, FunASR Python, or a remote API.

Rules

Read and follow dependency-manager before running commands.

  • Run .ai/audio-to-text/transcribe.py through the default python manifest entry. Do not invoke the native binary by hand.
  • Let the wrapper try funasr-llamacpp-gpu first with the Vulkan backend and fall back to funasr-llamacpp-cpu. Do not use CUDA, FunASR Python, PyTorch, ONNX, or host tools.
  • Load sensevoice-small-f16.gguf and fsmn-vad.gguf only from the sensevoice-small-gguf manifest model directory.
  • Process exactly one audio file per invocation. Repeat the command for multiple files.
  • Require uncompressed PCM16 WAV input. The current native runtime accepts common sample rates and channel layouts. For other containers, first chain audio-to-wav.
  • Always pass --vad for long-audio segmentation.
  • Produce plain UTF-8 transcription text only.
  • Never overwrite or modify the source audio.

Usage

Plain transcription:

.dependency/python/python.exe .ai/audio-to-text/transcribe.py --audio C:\path\speech.wav

Do not write a transcript file. Return the text printed to stdout. See cli/audio-to-text.md for tests and copy-paste commands.

Use --cpu only when the user explicitly requests CPU or Vulkan troubleshooting requires bypassing GPU acceleration.

Report

Return only the transcript. For long audio, summarize instead of dumping the entire transcript unless asked.

Troubleshooting

  • Invalid WAV: convert to uncompressed PCM16 WAV before retrying.
  • Vulkan fallback: update the GPU driver and confirm it exposes a Vulkan device; use --cpu to bypass Vulkan deliberately.
  • Slow inference: F16 prioritizes fidelity and uses more memory than Q8.
  • Missing runtime: verify both .dependency/funasr-llamacpp-gpu/llama-funasr-sensevoice.exe and the CPU runtime exist.
  • Missing model: verify both GGUF files exist under .dependency/sensevoice-small-gguf/model/.

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.