Whisper Transcription

Local speech-to-text using OpenAI Whisper. No API key needed, audio stays on your server.

tested

The Story

Enabling voice messages to AI agents without external API dependencies. Whisper was added because Brian realized that voice-first workflows break when the agent can't actually hear what you're saying. OpenClaw doesn't transcribe audio by default - Whisper fixes that.

Local speech-to-text using OpenAI Whisper - no API key needed, audio stays on your server.

The Problem

OpenClaw doesn't transcribe audio by default. When users send voice messages:

• They appear as audio attachments
• No transcript is generated
• The agent can't read the content

This breaks voice-first workflows. If you want to just talk instead of type, tough luck.

The Solution

Enable OpenAI Whisper for audio transcription in OpenClaw. Whisper is an open-source speech-to-text model that runs locally - no external API calls, no costs per minute, no data leaving your server.

How It Works

1. Voice message is captured in the conversation
2. Whisper CLI processes the audio locally
3. Transcript is returned as text
4. Agent can read and respond to the content

No data leaves your server. No API costs. Just local transcription.

Setup

Whisper is installed as a skill in DEWER:

openclaw skills install chiptrack

Once installed, voice messages are transcribed automatically when sent to DEWER.

Benefits

• No API costs - Uses local Whisper model
• Privacy - Audio stays on your server
• Reliability - No external service dependencies
• Speed - Local processing is fast

Use Cases

Enables hands-free interaction with DEWER:

• Mobile users who prefer voice
• Long-form input (speaking is faster than typing)
• Accessibility - helps users who prefer voice
• Capturing ideas while on the go

Ideas for Refinement

• Add speaker identification for multi-person audio
• Support for longer recordings
• Punctuation and formatting improvements
• Multiple language support