Expand your live video accessibility globally with near real-time multilingual subtitles. Translate speech into over 200 languages using AI-powered transcription and translation without manual intervention.
Overview
This Guidance demonstrates how to deliver real-time multilingual subtitles for live video content without adding streaming latency or compromising viewer experience, using AWS Elemental Media Services for video processing, Amazon Transcribe for speech-to-text conversion, and Amazon Bedrock with Nova Foundation Models for contextually aware translations across multiple languages. The live stream flows through AWS Elemental MediaConnect to AWS Elemental MediaLive for adaptive bitrate transcoding, while a parallel path sends audio to Amazon Transcribe and Amazon Bedrock for translation into 200+ languages. AWS Fargate containers manage the transcription and translation workload, storing time-synced subtitles in Amazon DynamoDB, while Amazon CloudFront with Lambda@Edge aligns timestamps to video playback and delivers the final subtitle files to viewers. You can expand into global markets while maintaining perfect synchronization between spoken content and subtitles, with serverless design that automatically scales from small webinars to millions of concurrent viewers at pay-as-you-use cost efficiency.
Benefits
Reach audiences in 200+ languages
Scale live events without infrastructure overhead
Deploy a serverless, container-based subtitle pipeline that scales automatically across multiple simultaneous live events. Reduce operational burden while maintaining time-synced subtitle delivery.
Accelerate subtitle delivery with parallel processing
Generate synchronized multilingual subtitles in near real-time by decoupling video transcoding from transcription and translation. Deliver precise, time-aligned captions to viewers with minimal latency.
How it works
This architecture diagram illustrates how auto multilingual subtitles for live video streams can be generated in near real-time on AWS.
Download the architecture diagram.
Step 1
A live stream is pushed to AWS Elemental MediaConnect from the source.
One output from AWS Elemental MediaConnect is pushed to AWS Elemental MediaLive for stream transcoding into an Adaptive Bitrate (ABR) stream.
Another output is pushed for transcription to Amazon Transcribe and then further translated using Amazon Bedrock and Amazon Nova Foundation Models, which supports 200+ languages and delivers frontier intelligence with industry-leading price-performance. This process runs in AWS Fargate containers, enabling easier scaling and management for multiple simultaneous live events.
The output of the transcription and translation process is segmented by an internal application, which is running in AWS Fargate, as a time-synced subtitle and stored in Amazon DynamoDB.
The ABR transcode, including blank WebVTT subtitle file, generated by AWS Elemental MediaLive is pushed to AWS Elemental MediaPackage via Amazon CloudFront with Lambda@Edge which adds timestamps, aligning to the video, in these WebVTT files.
Requests for media files are directed by Amazon CloudFront based on URL patterns to AWS Elemental MediaPackage.
Subtitle WebVTT file requests from users are directed by Amazon CloudFront to AWS Lambda via Amazon API Gateway.
AWS Lambda fetches the time-stamped subtitle file from AWS Elemental MediaPackage and then fetches the appropriate subtitles from Amazon DynamoDB. The function modifies the original subtitle file to insert the correct subtitles and finally returns the updated subtitle WebVTT file to the client.
Deploy with confidence
Everything you need to launch this Guidance in your account is right here.
Let's make it happen
Ready to deploy? Review the sample code on GitHub for detailed deployment instructions to deploy as-is or customize to fit your needs.
Related content
Scale global live reach with AWS powered real-time WebVTT multilingual subtitling
Learn how to implement real-time multilingual subtitling for live video streams using AWS services.