Audio Transcription WhatsApp Bot

Flask service that receives WhatsApp audio messages via Twilio and replies with a transcription using OpenAI Whisper.

Why

WhatsApp voice notes are slow to skim. This bot turns them into text.

Features

• Loads Whisper model once at startup
• Validates Twilio signatures
• Supports WhatsApp voice notes (OGG/Opus) and general audio/*
• Sync or async replies via env flag
• Health endpoint at /healthz

Requirements

• Python 3.10+
• ffmpeg installed on system path
• Twilio WhatsApp Sandbox or Business API

Install

pip install -r requirements.txt

Install ffmpeg if you don't have it:

brew install ffmpeg

Configure

Create a .env file:

ACCOUNT_SID=ACxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
AUTH_TOKEN=your_auth_token
FROM=whatsapp:+1415xxxxxxx
MODEL_NAME=small
PORT=5000
DEBUG=false
ASYNC_REPLY=false

Alternatively, copy and edit env.sample to .env.

Notes:

• FROM can be configured with or without the whatsapp: prefix; the app handles both.
• MODEL_NAME options: tiny, base, small, medium, large (trade speed vs accuracy).
• ASYNC_REPLY=true returns immediately and sends the transcription in a second message.

Run

python main.py

Expose locally for Twilio callbacks:

ngrok http 5000

Set your Twilio WhatsApp sandbox Inbound Webhook URL to:

POST https://<your-ngrok-domain>/whatsapp

Usage

• Send a voice note or audio file to your Twilio WhatsApp number
• You will receive the transcription back

Notes on Cost

Each audio incurs costs for messaging and compute. Whisper model size affects speed and cost; smaller models are faster and cheaper to run.