SuperAPI AI Directory

audio-transcription

General & other

View repository

Updated 10-15 21:41

About this skill

Audio Transcription

By default, use this Skill's built-in local-transcribe to run Qwen and MOSS, executed sequentially locally, fully offline, with results kept separately. The local offline path only supports macOS Apple Silicon (arm64).

Routing

  1. Multiple media items are completed one by one in the order given by the user; the next one starts only after the previous command exits, without concurrently launching Agents, processes, or online tasks.
  2. For each media item, first run scripts/local-transcribe plan "<media>" --json。
  3. SETUP_REQUIRED: read setup.md. Install only after the user explicitly requests setup. When the user explicitly requests an external drive, a persistent runtime, or a persistent transcription directory, first run scripts/manage-runtime storage status --json; after it confirms four independent directories, run storage configure, then continue setup. This configuration only saves paths; it does not migrate existing data, nor does it constitute upload authorization.
  4. Normal transcription: run scripts/local-transcribe "<media>" --json, then read the artifacts pointed to by the manifest.
  5. When the user explicitly requests Doubao, online, cloud, or permits uploading: read online-transcription.md. "Dual-model, parallel, cross-validation" alone does not constitute upload authorization.
  6. When multiple results need to be checked: read transcript-review.md。
  7. When the next_actions given by the CLI cannot resolve a failure, read troubleshooting.md。

Source of truth

Parameters, default values, side effects, output protocols, error codes, and recovery actions are governed by each script's --help、help <topic> and --json ; do not guess from this document or the reference.

For platform limitations, online uploads, and third-party licenses, see setup.md、online-transcription.md and NOTICE.md。

Do not rewrite or delete input files. Do not stuff the full verbatim transcript into the reply; unless the user requests it, returning the mode, status, artifact paths, and important points of doubt is sufficient.