VideoTranslate Prototype

Upload an MP4/MOV to S3 and EventBridge triggers AWS Batch to generate Whisper subtitles (and optional translation) — designed for long videos beyond Lambda limits.

What it is

VideoTranslate is an automation tool that turns uploaded videos into usable text artifacts.

It’s designed specifically for the “long video” case where AWS Lambda’s 15-minute limit becomes a constraint.

1. You upload a video file (.mp4 / .mov) to an S3 bucket.
2. An EventBridge rule detects the new object under a target prefix (e.g. raw/).
3. AWS Batch runs a containerized job on EC2:
   • downloads the video
   • uses Whisper to produce subtitle segments
   • optionally uses Amazon Translate for zh-TW
4. The job uploads .srt back to S3 and moves the original video to a processed/ prefix.
5. A DynamoDB record is written to prevent duplicate processing.

Why it matters

Transcripts unlock:

• Searchable knowledge (index transcripts into a vector DB)
• Faster summarization and note taking
• Subtitle generation and localization

Architecture sketch

• S3 as ingestion (raw/) and output (translated/..., processed/)
• EventBridge to trigger jobs on S3 object creation
• AWS Batch + EC2 for long-running compute
• Whisper for speech-to-text subtitles
• Amazon Translate (optional) for localization
• DynamoDB for idempotency (don’t redo the same video/language combo)

Demo

See the simulated pipeline UI on the Lab page.

Notes

I also experimented with a Lambda-based version for short clips, but the “main” design goal here is reliability for longer media — so Batch is the default architecture.