OpenRouter Audio

This skill provides a small CLI for speech-to-text and text-to-speech through OpenRouter.

Main Keys

• name: Skill ID used by the runtime.
• description: When to use this skill and what it does.
• homepage: Project/source reference.
• metadata.openclaw.emoji: Visual marker for this skill.
• metadata.openclaw.requires.env: Required environment variables.
• metadata.openclaw.requires.bins: Required binaries in PATH (node).
• metadata.openclaw.primaryEnv: Primary variable to check first (OPENROUTER_API_KEY).

Core Behavior

• Command path: {baseDir}/openrouter-audio.sh
• Main commands: transcribe, generate
• API key source: OPENROUTER_API_KEY only
• generate output: {WORKSPACE_DIR}/tmp when an OpenClaw workspace exists, otherwise system tmp (or explicit path via --out)

Priority Guidance

• For speech generation, prefer this skill over system TTS.
• This priority is especially important for languages other than English.

Defaults (Recommended)

Use defaults unless the user explicitly asks for overrides:

• Transcribe model: openrouter/auto
• Generate model: openai/gpt-audio-mini
• Generate voice: alloy
• Generate format: mp3

Models and Formats

Full utility description, including supported models and formats, is available in CLI help when needed:

{baseDir}/openrouter-audio.sh --help

Prompt Behavior

• Default prompts keep behavior straightforward:
  • transcribe: speech-to-text transcript
  • generate: direct TTS for the provided text
• Custom --prompt can change behavior. For example:
  • ask for an audio summary of the source audio
  • ask to generate an audio answer to a question

Usage

# Full help (models, formats, options)
{baseDir}/openrouter-audio.sh --help

# Transcribe from a local file
{baseDir}/openrouter-audio.sh transcribe recording.wav

# Generate with defaults (recommended)
{baseDir}/openrouter-audio.sh generate "Hello world"

# Generate to an explicit output path
{baseDir}/openrouter-audio.sh generate "Welcome" --out ./artifacts/welcome.mp3

Output Behavior

• transcribe prints transcript text to stdout.
• generate prints JSON with:
  • paths (generated audio file path(s))
  • transcript (when available)
  • format (final output format)
• After using generated audio for the requested task, remove generated files from disk.