Audio Transcription Service

A serverless audio transcription service built on AWS using OpenAI's Whisper model. This cloud-native application automatically transcribes audio files uploaded by users, providing job tracking and seamless client interaction through a REST API.

What This Application Does

The Audio Transcription Service is a fully managed, serverless solution that:

• Accepts audio file uploads through a secure REST API
• Automatically transcribes audio using OpenAI's Whisper model running in AWS Lambda containers
• Tracks job progress with real-time status updates stored in DynamoDB
• Provides secure download links for completed transcriptions
• Scales automatically based on demand without manual intervention
• Manages the entire workflow from upload to transcription completion

Key Features

• ✅ Serverless Architecture - No infrastructure management required
• ✅ Secure File Handling - Pre-signed URLs for secure uploads and downloads
• ✅ Job Tracking - Real-time status monitoring with unique job IDs
• ✅ Auto-scaling - Handles varying workloads automatically
• ✅ Cost-effective - Pay only for what you use
• ✅ High Availability - Built on AWS managed services

Repository Structure

cloud_computing_project/
├── src/
│   ├── client-CLI/
│   │   └── client.py              # Python CLI client for testing the service
│   └── lambda-functions/
│       ├── api-lambda.py          # API Gateway handler Lambda function
│       └── whisper-lambda/
│           ├── app.py             # Whisper transcription Lambda function
│           ├── Dockerfile         # Container image for Whisper Lambda
│           └── requirements.txt   # Python dependencies for Whisper
├── tests/                         # Performance tests code and results
└── README.md                      # This file

Folder Descriptions

/src/client-CLI/

Contains a Python command-line client that demonstrates how to interact with the transcription service:

• client.py - Full-featured CLI client that can upload audio files, poll for job status, and download completed transcriptions

/src/lambda-functions/

Contains the AWS Lambda functions that power the service:

• api-lambda.py - Handles HTTP requests from API Gateway, manages job creation and status retrieval
• whisper-lambda/ - Container-based Lambda function that performs the actual audio transcription using OpenAI's Whisper model

Architecture

The service follows a serverless architecture pattern using AWS managed services:

flowchart TB
 subgraph subGraph0["S3 Bucket Structure"]
        S3Audio["audio/<br>Uploaded audio files"]
        S3Trans["transcripts/<br>Generated transcriptions"]
  end
 subgraph subGraph2["AWS Services"]
        API["API Gateway"]
        Lambda1["API Handler Lambda<br>api-lambda.py"]
        Lambda2["Whisper Lambda<br>whisper-lambda/app.py"]
        DDB[("DynamoDB<br>TranscriptionJobs")]
        S3["S3 Bucket"]
        CloudWatch["CloudWatch<br>Monitoring &amp; Logs"]
  end
    Client["Client Application"] --> API
    API --> Lambda1
    Lambda1 --> DDB & S3
    S3 -- S3 Event Trigger --> Lambda2
    Lambda2 --> S3 & DDB
    S3 --> S3Audio & S3Trans
    Lambda1 -.-> CloudWatch
    Lambda2 -.-> CloudWatch

    style API fill:#f3e5f5
    style Lambda1 fill:#fff3e0
    style Lambda2 fill:#fff3e0
    style DDB fill:#e8f5e8
    style S3 fill:#fff8e1
    style CloudWatch fill:#f1f8e9
    style Client fill:#e1f5fe

Loading

Architecture Flow

1. Job Creation: Client sends POST request to /upload endpoint

   • API Lambda generates unique jobId and pre-signed S3 upload URL
   • Job status saved as "UPLOADING" in DynamoDB

2. File Upload: Client uploads audio file directly to S3 using pre-signed URL

   • File stored in audio/ folder with jobId tag

3. Transcription Processing: S3 upload triggers Whisper Lambda

   • Downloads audio file from S3
   • Processes with OpenAI Whisper model in container
   • Saves transcript to transcripts/ folder
   • Updates job status to "COMPLETED" in DynamoDB

4. Result Retrieval: Client polls /status/{jobId} endpoint

   • Returns job status and download URL when ready
   • Pre-signed URL provides secure access to transcript

Core Components

• Amazon S3: Stores audio files and generated transcripts
• AWS Lambda: Serverless compute for API handling and transcription processing
• API Gateway: REST endpoint management with CORS support
• DynamoDB: Job tracking and status management
• CloudWatch: Monitoring, logging, and performance metrics

---

This project is part of the Cloud Computing course final project, demonstrating serverless architecture, auto-scaling, and performance evaluation on AWS.