Welcome to Silicon Overdrive!

Linkedin Youtube

Welcome to Silicon Overdrive!

Making Accessibility Content Creation Scalable with Generative AI

1. About the Customer

Phonetik is a South African media technology company focused on making video content more accessible. Its AI-powered platform helps content creators, broadcasters, and media distributors generate accessibility assets such as closed captions, hearing-impaired subtitles, dialogue sheets, and audio description tracks.

As demand for accessible content continues to grow globally, Phonetik set out to enhance its platform’s ability to process content at scale while maintaining the quality and accuracy required by professional media customers.

2. The Challenge

While Phonetik’s platform could successfully transcribe and diarise audio, identifying exactly who was speaking remained a manual process. Human editors were required to review content and assign dialogue to the correct on-screen speaker before accessibility assets could be finalized.

This created a significant operational bottleneck. As content volumes increased, the company faced the prospect of scaling editorial resources alongside customer growth, limiting its ability to efficiently support larger content libraries, enterprise customers, and broadcast-scale workloads.

At the same time, growing accessibility requirements across global markets were increasing demand for high-quality, scalable accessibility solutions. Phonetik needed a way to automate speaker attribution without disrupting its existing platform or customer experience.

3. The Solution

Silicon Overdrive partnered with Phonetik to design and implement a generative AI-powered speaker attribution solution on AWS.

The solution combines Amazon SageMaker and Amazon Bedrock AgentCore to automate the process of determining who is speaking within video content. Using an event-driven architecture built on AWS, the platform analyzes transcription data, speaker diarisation results, face detection information, and active speaker detection outputs to make intelligent attribution decisions.

Amazon Bedrock AgentCore acts as the reasoning layer, evaluating confidence levels and applying business rules to handle complex scenarios such as:

  • Multiple speakers appearing simultaneously
  • Off-screen dialogue
  • Unknown speakers
  • Low-confidence attribution events
  • Overlapping conversations

The solution integrates directly into Phonetik’s existing workflow, allowing accessibility outputs to be generated automatically while preserving the company’s existing customer experience and operational processes.

By leveraging managed AWS services, Phonetik also gained a highly scalable and resilient platform capable of supporting future growth without increasing operational complexity.

4. The Results

The implementation delivered measurable improvements across both automation and platform performance.

1. 90% Speaker Attribution Accuracy

The solution achieved 90% attribution accuracy across Phonetik’s validation assets, exceeding the agreed target and demonstrating the effectiveness of combining machine learning with generative AI reasoning.

2. 93% End-to-End Processing Success Rate

The automated processing pipeline achieved a 93% completion rate, successfully processing content from ingestion through to final attribution output with minimal manual intervention.

3. Reduced Manual Effort

By automating speaker attribution, Phonetik significantly reduced the manual editorial work previously required to prepare captions, dialogue sheets, and accessibility outputs.

4. Improved Scalability

The new architecture enables Phonetik to process larger volumes of content more efficiently, supporting future growth without requiring a proportional increase in editorial resources.

5. Strong Foundation for Continued Innovation

The deployment established a scalable AI foundation that can continue to evolve as additional content types, attribution scenarios, and accessibility requirements emerge.

4. Customer Impact

With automated speaker attribution now embedded within its accessibility workflow, Phonetik is better positioned to help media organizations create accessible content at scale. The solution enables faster content processing, reduced operational overhead, and a stronger platform for supporting the growing demand for accessible media experiences.

Technologies Used: Amazon Bedrock AgentCore, Amazon SageMaker, Amazon Transcribe, Amazon Rekognition, AWS Lambda, Amazon S3, Amazon SQS, Amazon DynamoDB, Amazon RDS for PostgreSQL, Amazon EventBridge.

5. What the customer has to say:

Francois Schreuder, CEO, Phonetik said, “On the security and compliance side, Trend Vision One compliance validation scored 95–97% across the AWS Well-Architected Framework, NIST Cybersecurity Framework, ISO 27001:2022, HIPAA, PCI DSS and GDPR, meeting the proposal’s 95%+ GDPR/CREM success metric.”

Visit their website

If you liked this, you’ll love these...

Oye Platform Solutions Limited

Learn how Oye Platform Solutions partnered with Silicon Overdrive to migrate Songa Na Oye, its lending and insurance platform for boda boda operators, from Azure to a resilient, best-practice AWS environment.

We use cookies to track visitors, measure ads, ad campaign effectiveness and analyze site traffic. We may also share information about your use of our site with 3rd parties. For more info, see, our Cookies Policy, our Privacy Notice. By clicking “Accept All” you agree to the storing of all cookies on your device. In case you don’t choose one of these options and use our website, we will treat it as if you have accepted all cookies.

Accept All

We use cookies to track visitors, measure ads, ad campaign effectiveness and analyze site traffic. We may also share information about your use of our site with 3rd parties. For more info, see, our Cookies Policy, our Privacy Notice. By clicking “Accept All” you agree to the storing of all cookies on your device. In case you don’t choose one of these options and use our website, we will treat it as if you have accepted all cookies.

Accept All