whisper-cpp for Openclaw

A high-performance local speech-to-text engine for OpenClaw using whisper.cpp for private, offline transcription.

truenight
v1.0.2
Feb 27, 2026
0
1.2k
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install whisper-cpp

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install whisper-cpp using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is whisper-cpp?

The whisper-cpp skill brings high-performance, local speech-to-text capabilities to your AI agent environment. By integrating the optimized whisper.cpp engine, this skill allows for the transcription of voice notes and audio files without sending data to external providers or incurring API costs. It is an essential component for users building privacy-first Openclaw Skills that need to process inbound audio data directly on their own hardware.

This skill is particularly effective for handling Telegram voice notes, leveraging ffmpeg for seamless format conversion and whisper.cpp for efficient inference. Because it supports various ggml model sizes, users can balance speed and accuracy based on their specific server or desktop specifications, making it one of the most versatile Openclaw Skills for media processing.

whisper-cpp Use Cases

  • Transcribing inbound Telegram voice notes privately without third-party API leaks.
  • Running speech-to-text workflows in air-gapped or offline environments.
  • Reducing costs for high-volume audio processing by eliminating per-minute transcription fees.
  • Implementing custom voice-controlled triggers for local automation scripts.

How whisper-cpp Works

  1. The installation process compiles the whisper.cpp binaries locally on your machine to ensure hardware optimization.
  2. User-selected ggml models are downloaded and stored in a local cache directory for persistent use.
  3. A dedicated wrapper script is installed to act as the bridge between OpenClaw and the whisper engine.
  4. The skill patches the OpenClaw configuration to register the local STT tool as the primary audio processor.
  5. When an audio file is received, it is converted to a compatible WAV format and processed by the engine to return text to the AI agent.

whisper-cpp Setup

To install the whisper-cpp skill, ensure you have build tools, curl, and ffmpeg installed, then execute the following commands from the skill directory:

bash scripts/install_whisper_cpp.sh
bash scripts/download_models.sh
bash scripts/install_wrapper.sh
bash scripts/patch_openclaw_audio.sh

You can customize which models are downloaded by passing them as arguments:

bash scripts/download_models.sh tiny base small

whisper-cpp Data Schema & Taxonomy

The skill manages its assets and configurations using the following structure:

Component Path Description
Models ~/.cache/whisper Stores ggml model files (e.g., base, small, large-v3)
Binaries ~/.local/bin/ Contains the openclaw-whisper-stt executable and whisper-cli
Config OpenClaw Integrated via the tools.media.audio configuration keys
Wrapper bin/ Source for the OpenClaw wrapper script

whisper-cpp Advanced Features

  • Support for the full range of ggml models, including large-v3 for professional-grade accuracy.
  • Environment variable support for forcing specific languages using OPENCLAW_WHISPER_LANG.
  • Ability to switch between models on-the-fly via the OPENCLAW_WHISPER_MODEL variable.
  • Automated OGG/Opus to WAV conversion pipeline optimized for Telegram integration.
  • Built-in troubleshooting tools for checking path resolution and configuration status.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*