Create subtitles for your video-on-demand content to reach more audiences, globally
Publication date: November 2021 (last update: October 2024)
Content Localization on AWS helps extend the reach of your video-on-demand (VOD) content
by efficiently creating accurate multi-language subtitles using Amazon Web Services (AWS) Artificial
Intelligence (AI) services. You can make manual corrections to the automatically created
subtitles and use advanced AWS AI service customization features to improve the results of the
content automation. This solution is built on Media Insights
on AWS
Localization is the process of taking video content that was created for audiences in one geography and transforming it to make it relevant and accessible to audiences in a new geography. Creating alternative language subtitle tracks is central to the localization process. With this solution, content owners can use a guided experience for automatically generating and correcting subtitles for videos in multiple languages using AWS AI services. Corrections made by solution users can be used as feedback to customize the results of AWS AI services for future workflows. This type of AI/ML (Machine Learning) workflow, which incorporates user corrections is referred to as human-in-the-loop.
Content localization workflows can make use of advanced customization features provided by Amazon Transcribe and Amazon Translate:
-
Amazon Transcribe custom vocabulary - you provide Amazon Transcribe with a list of terms that are specific to your content and how you want the terms to be displayed in transcripts.
-
Amazon Transcribe custom language models - you provide Amazon Transcribe with text data (training data) to improve transcription accuracy for industry-specific terms, acronyms, and other phrases that it might not otherwise recognize. This becomes a text corpus
list of terms that are specific to your content and how you want the terms to be displayed in transcripts. -
Amazon Translate custom terminologies - you provide Amazon Translate with a list of terms or phrases in the source language content and specify how you want them to appear in the translated result.
-
Amazon Translate parallel data for active custom translation - you provide Amazon Translate with a list of parallel phrases: the source language and the phrase translated the way you want it. The parallel data customizes Amazon Translate models so they create more contextual translations based on the sample you provide.
Application users can manually correct the results of the automation at different points in the automated workflow and then invoke a new workflow to include their corrections in downstream processing. Corrections are tracked and can be used to update Amazon Transcribe custom vocabularies and Amazon Translate custom terminologies to improve future results.
Why use customizations and human-in-the-loop?
Automating the creation of translated subtitles using AI/ML helps to speed up the process of localization for your content, but there are still challenges to achieve the level of accuracy that is required for specific use cases. With natural language processing, many aspects of the content itself may determine the level of accuracy AI/ML analysis is capable of achieving. Some content characteristics that can impact transcription and translation accuracy include: domain specific language, speaker accents and dialects, new words recently introduced to common language, the need for contextual interpretation of ambiguous phases, and correct translation of proper names. AWS AI services provide a variety of features to help customize the results of the machine learning to specific content. Therefore, the workflow in this application seeks to provide users with a guided experience to use these customization features as an extension of their normal editing workflow.
This implementation guide discusses architectural considerations and configuration steps
for deploying the Content Localization in AWS solution in the AWS Cloud. It includes links
to an AWS CloudFormation
The guide is intended for IT infrastructure architects and developers who have practical experience architecting in the AWS Cloud.
Use this navigation table to quickly find answers to these questions:
| If you want to . . . | Read . . . |
|---|---|
| Know the cost for running this solution The one-time estimated cost for processing a 30-minute video with five languages using the default settings in the US East (N. Virginia) Region is USD $5.64 per month. |
Cost |
| Understand the security considerations for this solution. | Security |
| Know how to plan for quotas for this solution. | Quotas |
| Know which AWS Regions are supported for this solution. | Supported AWS Regions |
| View or download the AWS CloudFormation template included in this solution to automatically deploy the infrastructure resources (the “stack”) for this solution. | AWS CloudFormation template |
| Access the source code and optionally use the AWS Cloud Development Kit (AWS CDK) to deploy the solution. | GitHub
repository |
