whisper-stt

name	whisper-stt
description	Speech-to-Text transcription using Whisper. Transcribes voice messages from Odoo attachments to text. Supports local Faster Whisper (no API key) and OpenAI Whisper API fallback. Use when user sends a voice note or audio attachment.
homepage	https://github.com/nicolasramos/odooclaw
metadata	{"openclaw":{"emoji":"🎤","requires":{"env":["ODOO_URL","ODOO_DB","ODOO_USERNAME","ODOO_PASSWORD","OPENAI_API_KEY"]}}}

Whisper STT Skill

Overview

This skill transcribes voice messages from Odoo attachments to text using Whisper. It supports two methods:

1. Faster Whisper (local) - No API key needed, runs on CPU
2. Whisper API (OpenAI) - More accurate, requires OPENAI_API_KEY

The skill automatically tries Faster Whisper first, then falls back to API if needed.

Quick Reference

Use Case	Tool
Transcribe voice note	whisper-transcribe
Check available methods	whisper-list-methods

Tools

whisper-transcribe

Transcribes an audio attachment from Odoo.

Parameters:

• attachment_id (required): The Odoo ir.attachment ID to transcribe

Returns: Transcribed text, detected language, and method used

whisper-list-methods

Lists available transcription methods and their status.

Usage Examples

Example 1: Transcribe a Voice Message

When user sends a voice note, the webhook includes voice_attachments with the attachment ID.

Action:

{
  "name": "whisper-transcribe",
  "arguments": {
    "attachment_id": 1234
  }
}

Result:

{
  "success": true,
  "text": "Hola, me gustaría saber el estado del pedido S00001",
  "language": "es",
  "method": "faster_whisper"
}

Example 2: Full Workflow

1. User sends voice note → webhook includes voice_attachments: [{id: 1234, name: "voice.mp3"}]
2. Use whisper-transcribe to get text
3. Use odoo-mcp tools to query the data user is asking about
4. Reply with text or use edge-tts-synthesize for voice response

Integration with Odoo

Webhook Payload Changes

When a voice message is received, the webhook now includes:

{
  "message_id": 100,
  "model": "discuss.channel",
  "res_id": 5,
  "author_id": 3,
  "author_name": "John Doe",
  "body": "\n🎤 [Nota de voz: voice_20240315.mp3 (ID: 123)]\n",
  "is_dm": true,
  "voice_attachments": [
    {
      "id": 123,
      "name": "voice_20240315.mp3",
      "mimetype": "audio/mp4"
    }
  ],
  "attachments": []
}

Detecting Voice Messages

Check for voice_attachments array in the webhook payload. Each voice attachment has:

• id: The attachment ID to pass to whisper-transcribe
• name: Original filename
• mimetype: Audio MIME type

Environment Variables

Variable	Required	Description
ODOO_URL	Yes	Odoo server URL
ODOO_DB	Yes	Odoo database name
ODOO_USERNAME	Yes	Odoo username
ODOO_PASSWORD	Yes	Odoo password or API key
OPENAI_API_KEY	No	For Whisper API fallback

Configuration

Using Faster Whisper Only (No API Key)

# docker-compose.yml
environment:
  - OPENAI_API_KEY=  # Not needed

The skill will use Faster Whisper (local) for all transcriptions.

Using Whisper API (More Accurate)

environment:
  - OPENAI_API_KEY=sk-...

The skill will use Whisper API for better accuracy.

Using Both (Recommended)

environment:
  - OPENAI_API_KEY=sk-...

1. Faster Whisper is tried first (free, local)
2. If it fails, Whisper API is used as fallback

Supported Audio Formats

• MP3
• MP4
• WAV
• OGG
• WebM
• M4A

Limitations

• Faster Whisper: Uses "tiny" model by default for speed. Larger models = more accurate but slower.
• Audio length: No hard limit, but very long audio may timeout.
• API cost: Whisper API costs ~$0.006/minute (as of 2024).

Best Practices

1. Check for voice first: Always check voice_attachments in webhook payload
2. Confirm transcription: Show user the transcribed text before processing
3. Handle errors gracefully: If transcription fails, ask user to send text instead
4. Language: Faster Whisper auto-detects language; Whisper API does too