Voice Enrich

AI-only enrichment for MindRoom voice messages.

When a transcribed voice message arrives, this plugin silently injects a note into the agent prompt explaining that the text came from speech recognition and may contain transcription mistakes. The user-facing Matrix event body is never modified.

Features

• Detects voice-origin messages using mindroom.constants.VOICE_PREFIX (🎤 )
• Injects AI-only metadata via MessageEnrichContext.add_metadata
• Warns about common speech-to-text failure modes such as homophones, proper nouns, and rare technical terms
• Suggests re-transcribing the original audio attachment when the transcript looks garbled
• Zero-config, hooks-only behavior
• Lightweight execution with no network or file I/O in the hook path

How It Works

1. MindRoom transcribes an incoming voice message and prefixes the visible body with 🎤 .
2. On message:enrich, the voice-enrich hook checks whether the incoming body starts with VOICE_PREFIX.
3. If it does, the hook adds a volatile voice_transcription metadata note to the prompt context.
4. The agent sees the note during reasoning, but the Matrix message body shown to the user stays unchanged.

Hooks

Hook	Event	Purpose
voice-enrich	message:enrich	Add voice transcription guidance when the message body starts with 🎤

Install

Vendor this plugin with the MindRoom CLI:

mindroom plugins install voice-enrich-plugin

Then reference it from config.yaml:

plugins:
  - path: plugins/voice-enrich-plugin

Update to the latest commit later with:

mindroom plugins update voice-enrich-plugin

The command pins the exact installed commit in .mindroom-plugin.lock.json and strictly validates the plugin before activating it. For a manual checkout instead, see Setup below.

Setup

1. Copy this plugin to ~/.mindroom/plugins/voice-enrich.
2. Add the plugin to config.yaml:

   plugins:
     - path: plugins/voice-enrich
   

3. Restart MindRoom.

No agent tools or plugin settings are required. This plugin is hooks-only.