Die Oberfläche ist lokalisiert; Skill-Beschreibungen und -Dokumente werden per KI automatisch in Ihre Sprache übersetzt. Code, Befehle, Pfade und Fachbegriffe bleiben im Original, maßgeblich ist das Dokument des Autors.
Skills entdecken
Produktivität · SKILL.md

nodejs-server

nodejs-server

Use Porcupine from Node.js server-side JavaScript or TypeScript for file and microphone wake-word detection.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill-Dokumentation

GitHub

name: nodejs-server description: "Use Porcupine from Node.js server-side JavaScript or TypeScript for file and microphone wake-word detection." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Porcupine Node.js Server Skill

Use this sub-skill when the task is to run Porcupine from Node.js 18+ in a server, CLI, Electron main-process, or other non-browser JavaScript/TypeScript runtime.

Route here for

  • Installing and importing @picovoice/porcupine-node.
  • Constructing new Porcupine(accessKey, keywords, sensitivities, options).
  • Choosing built-in keyword enum values versus custom .ppn keyword file paths.
  • Processing 16 kHz, 16-bit, mono PCM frames with process(frame) and calling release().
  • Listing inference devices with Porcupine.listAvailableDevices().
  • Building file-based WAV scanning or microphone capture workflows in Node.js.
  • Diagnosing native .node package loading, AccessKey, frame-shape, sensitivity, and lifecycle failures.

Route elsewhere

  • Browser, WebAssembly, React hooks, or browser microphone work: ../web-and-react/SKILL.md.
  • Custom keyword training, language model .pv selection, platform-specific .ppn assets, and Picovoice Console/API limits: ../custom-keywords-and-assets/SKILL.md.
  • Python CLI/API workflows: ../python-and-cli/SKILL.md.
  • C, MCU, mobile, JVM, .NET, Flutter, or React Native SDKs: the relevant sibling sub-skill.
  • Cross-cutting install/AccessKey/platform issues shared by every SDK: ../../references/troubleshooting.md.

First reads

  1. references/nodejs-api-reference.md for API signatures, built-in keyword names, options, exported helpers, package/build facts, and validation rules.
  2. references/nodejs-workflows.md for file WAV and microphone workflow recipes.
  3. references/troubleshooting.md before changing code around native loading, AccessKey, keyword arrays, PCM frames, or cleanup.
  4. scripts/porcupine_node_file_template.js when a safe starting point for a Node file-detection helper is useful.

Operating guardrails

  • Treat @picovoice/porcupine-node as a Node-only native package; do not use it in browser bundles.
  • Do not hard-code AccessKeys or require them for help, keyword listing, or device-listing tasks. An AccessKey is required only when constructing a Porcupine engine for detection.
  • Prefer BuiltinKeyword.<NAME> values for built-in English wake words. Use custom .ppn paths only for trained or non-bundled keyword models, and route asset/training decisions to custom-keywords-and-assets.
  • Keep every audio frame at handle.frameLength, handle.sampleRate, mono, 16-bit linear PCM. Use Int16Array in new code.
  • Always release the Porcupine handle in a finally block or shutdown handler; release microphone recorders separately.
Gespeicherte Skills