Better accessibility & discoverability of a MOOC using AI

Edraak, an initiative of the Queen Rania Foundation, is the largest MOOC (Massive Online Open Course) in the Arabic language, providing learning opportunities to millions of users. Using Artificial Intelligence (AI) solutions in partnership with Daemon, Edraak are now enabling a wider range of learners than ever to discover and access learning content.

The challenge

As a learning platform and charitable initiative, Edraak is working on getting its services out to as many people as possible, including people who prefer reading or people from different language backgrounds such as English. Edraak also needs to ensure those services are accessible to all - including people with sensory impairments such as hearing loss.

During our discovery phase we identified automatic transcription and translation of video content as a low-cost way of leveraging AI to get a big positive impact on Edraak's learners. By providing content creators with AI-powered tools for making subtitles on their videos with minimal friction, users with hearing impairment or a preference for reading will be able to access the core video content of courses, and the resulting subtitles and transcripts will be made available in different languages. This will be for the benefit of users and to provide search engine optimisation for content outside of Arabic and showcase Edraak to the whole world.

Our Approach

We decided to leverage the mature tools of the Google Cloud Platform, where Edraak is currently hosted, particularly its Cloud Text-to-Speech and Cloud Translation tools. This puts the heavy computational lifting into the cloud and into infrastructure-free ‘best of breed’ cloud APIs, freeing up Edraak to concentrate on how to integrate and make our prototype work for them and their users.

Using these APIs behind the scenes we built our prototype directly into the Django-based Open EdX framework that Edraak works with. We created a back-end transcription and translation workflow that would automatically discover video content, work on processing videos on demand, and ultimately put the results back into the Django-managed application database for users to benefit from.

We created a web-based UI that allows creators to view a video and edit the automatically generated transcript online as subtitles; the results are again stored back to the OpenEdX database where they can be picked up by Edraak's users. The resulting transcripts and translations would then be made available for search engine optimisation and discovery.

Outcome

API

A Django API for Video Transcription/Translation, interfacing with:

The Open edX/Edraak primary database running on Cloud SQL, for accessing current course information and maintaining transcription state and transcripts.
A new Open edX/Edraak content database running on Cloud SQL for discovering and retrieving videos.
Cloud Storage for storing preprocessed audio.
Cloud Text-to-Speech for doing the grunt-work of transcription.
Cloud Translation API for converting from the source language (usually, but not always, Arabic) to any target language.

Front end

A React UI for accessing and playing videos, viewing the resulting transcripts, and editing them as subtitles.

Component relationship

Results

Now possible to easily create accessible content for the hearing impaired and others.
90-95% certainty on AI-generated transcriptions.
A quick and easy subtitle editor for content creators.
Improved site discovery and SEO results.