Audio/Video Processing Pipeline on AWS

This project outlines a modular, cloud-native workflow for automating audio and video media processing using AWS services. It handles media uploads, conversion, metadata extraction, and asset preparation for public delivery.

Note: This is a generic template. You’ll need to customize bucket names, IAM roles, CORS policies, lifecycle rules, file paths, and environment variables to match your environment and security posture.

---

🗂 Folder Structure Overview

• s3://<watch-bucket>/inputs/: Incoming upload directory (watch folder)
• s3://<public-bucket>/: Destination for processed media
  • /audiovideo/ — Transcoded media files
  • /closed-caption/ — VTT caption files
  • /json/ — Video metadata
  • /original/ — Raw uploaded files
  • /thumbnail/ — Splash images (JPG)

---

📥 Upload & Initial Trigger

1. Files are uploaded via a web interface (e.g., using plupload) to the inputs/ folder.
2. This triggers an AWS Lambda function (e.g., MediaLambdaHandler).
3. Determines file type (audio/video) and checks for multipart CHUNK uploads.
   • CHUNK files (e.g., .CHUNK0, .CHUNK1) are ignored until fully assembled.

---

🧠 Media Analysis

Once a complete file is detected:

• A custom-compiled binary (e.g., pymediainfo) analyzes the file.
• Extracts key metadata: media type, resolution, duration, etc.

---

🎧 Audio File Handling

• MP3 files are moved directly to the final media bucket (e.g., audiovideo/).
• No additional processing is required.

---

🎞 Video File Conversion

1. Metadata determines which resolutions to create (e.g., skip 1080p for a 720p source).
2. A template job configuration (e.g., job.json) is modified in-memory.
3. AWS MediaConvert is invoked to:
   • Create multiple resolution variants (360p, 480p, 720p, 1080p).
   • Append resolution to filenames (filename_1080p.mp4).
   • Store the files in the designated output folder.

---

🧾 Metadata & Thumbnails

• Video metadata is saved as a .json file with a matching name in the /json/ folder.
• A separate Lambda function (e.g., SaveVideoFrame) extracts a JPG frame (default at 7 seconds).
• The thumbnail is saved to the /thumbnail/ folder.

---

📝 Closed Caption Support

• Caption files uploaded by users are converted to .vtt if necessary.
• They are stored with no additional processing in /closed-caption/.

---

🧰 Codebase & Deployment Artifacts

Lambda Source Code

• MediaLambdaHandler.py — Main Lambda for processing uploads
• SaveVideoFrame.py — Lambda for thumbnail extraction
• job.json — Template for AWS MediaConvert job settings

Custom Binaries

• pymediainfo and ffmpeglayer should be compiled and packaged as Lambda layers compatible with your Python runtime (e.g., 3.7).
• A good starting guide:
  https://binx.io/blog/2017/10/20/how-to-install-python-binaries-in-aws-lambda/

---

⚙️ Lambda Configuration Notes

MediaLambdaHandler

• Environment Variables (set per deployment):
  • APPLICATION_NAME
  • DESTINATION_BUCKET
  • MEDIACONVERT_ROLE_ARN
• Basic Settings:
  • Runtime: Python 3.7+
  • Memory: 128 MB
  • Timeout: 2 minutes
  • S3 Trigger: Event type ObjectCreated with prefix inputs/

SaveVideoFrame

• Important: Thumbnail generation is memory-intensive.
• Settings:
  • Runtime: Python 3.7+
  • Memory: 3008 MB
  • Timeout: 30 seconds
  • Trigger: Upload to /original/ folder

---

📡 Monitoring & Notifications

• CloudWatch logs are enabled by default; all scripts use logger.info() for traceability.
• An EventBridge (CloudWatch Events) rule can be set up to notify an SNS topic when MediaConvert jobs complete.
  • Subscribers can include dev team members, automated systems, or webhooks.

---

🔒 Security & Access

⚠️ You are responsible for defining IAM roles, policies, and bucket configurations.

• Watchfolder S3 Bucket:

  • Should block public access.
  • Recommended: Lifecycle rule to delete old uploads after 24 hours.

• Public Delivery Bucket:

  • Should be read-accessible as needed (e.g., via CloudFront or direct links).
  • Configure CORS policy to allow cross-origin access as necessary.

---

📌 Deployment Tips

• Use Infrastructure-as-Code (e.g., CloudFormation, Terraform, CDK) to manage:
  • Lambda functions
  • S3 buckets and policies
  • MediaConvert roles and permissions
• Set up versioning on buckets if retaining media history is needed.
• Automate unit tests for metadata validation and file integrity.

---

🧪 Testing

For local testing:

• Use a tool like localstack to mock AWS services.
• Upload test media files to a local S3 bucket to simulate the full pipeline.

---

💬 Final Thoughts

This pipeline offers a flexible blueprint for media processing in AWS. It’s scalable, customizable, and designed for clarity. Just plug in your infrastructure, tweak the knobs, and let the automation do the heavy lifting.

---

🛠 To-Do / Customization Checklist

• Replace bucket names (<watch-bucket>, <public-bucket>)
• Replace placeholder ARNs and role names
• Configure lifecycle & CORS policies
• Compile and upload Lambda layers (pymediainfo, ffmpeglayer)
• Setup CloudWatch and SNS alerts for your team