Whisper Transcription Cloudflare Worker

Project Overview

This project implements an AI-powered audio transcription service using Cloudflare Workers AI and the Whisper-large-v3-turbo model. The application can transcribe audio files of various lengths by supporting chunk-based processing and leveraging Cloudflare's serverless infrastructure.

Features

• Automatic Speech Recognition (ASR) using OpenAI's Whisper model
• Supports large audio file transcription through intelligent chunking
• Cloudflare Workers deployment for scalable, low-latency transcription
• Configurable transcription parameters:
  • Language selection
  • Translation vs. transcription mode
  • Voice activity detection
  • Custom initial prompts

Prerequisites

• Cloudflare account
• Node.js (v18+ recommended)
• Wrangler CLI
• Basic JavaScript/TypeScript knowledge

Installation

1. Clone the repository:

git clone <your-repo-url>
cd whisper-transcription-worker

2. Install dependencies:

npm install

3. Configure Cloudflare credentials:

npx wrangler login

Configuration

Update wrangler.toml with:

compatibility_date = "2024-09-23"
nodejs_compat = true

[ai]
binding = "AI"

Local Development

Start the development server:

npx wrangler dev --remote

Deployment

Deploy your Worker:

npx wrangler deploy

Usage Example

// Sample API call configuration
const transcriptionOptions = {
  audio: base64EncodedAudio,
  task: "transcribe",
  language: "en",
  vad_filter: "false"
}

Supported Audio Formats

• MP3
• WAV
• Other formats supported by Cloudflare Workers AI