L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

cli-model-setup

cli-model-setup

Install SpeechRecognition, select optional extras, prepare CLI/model assets, and check runtime environments.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: cli-model-setup description: "Install SpeechRecognition, select optional extras, prepare CLI/model assets, and check runtime environments." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

CLI Model Setup

Use this sub-skill when the user asks how to install SpeechRecognition, choose optional extras, validate an environment, use the sprc command, prepare the Vosk model expected by the package, or understand the interactive module entry points.

Route quickly

  • Package identity and install facts: distribution name SpeechRecognition, import package speech_recognition, console script sprc, Python >=3.10. See optional dependencies.
  • CLI and model setup: use CLI reference before running sprc download vosk; that command downloads from the network and writes into the installed package's model directory.
  • Non-invasive validation: run the bundled environment checker to inspect the installed version, recognizer method wiring, sprc help, and optional import groups without downloading models or checking authentication.
  • Install/CLI failure diagnosis: see troubleshooting, especially the tqdm CLI import pitfall and Python 3.13+ compatibility-package issues.

Boundaries

  • Recognition method selection, parameters, authentication, and transcription behavior belong to recognition-engines.
  • Audio file loading, conversion, chunking, FLAC payloads, and AudioData manipulation belong to audio-data.
  • Microphone capture, PyAudio device selection, ambient-noise calibration, and python -m speech_recognition demo behavior belong to capture-listening.
  • Maintainer CI matrices, repository tests, lint/typecheck/rstcheck, and release tasks belong to repo-development.

Safe operating rules

  • Do not run sprc download vosk unless the user accepts network download, unzip, overwrite, and installed-package write side effects.
  • Do not run python -m speech_recognition as an automated check; it is an interactive microphone demo and can block or call a network recognizer.
  • Prefer targeted extras over installing every optional group. Install only the workflow the user needs, then verify with the bundled checker.
Skills enregistrées