Executive Summary

Our Customer, the largest media company in the Middle East and North Africa, was struggling with issues related to their AWS content localization solution used for automating transcription and translation processes. Initially, the customer weren't able to translate transcription files in VTT and SRT formats. To tackle these problems, we made improvements to the content localization solution, which now has the capability to handle workloads on transcribed files. Additionally, our client sought to improve transcription accuracy in both Arabic and Turkish languages. The upgraded solution incorporated tuned OpenAI's Whisper, which significantly improved the transcription accuracy, reaching up to 90% for both Arabic and Turkish.

Customer Challenge

Content Localization solution isn't developed to handle transcripted files in vtt or srt format directly. The customer were supposed to upload the main video each time alongside with the transcription to translate the transcription. The challenge lies in integrating the existing solution with these file formats seamlessly. This requires modifications to the workflow and ensuring compatibility without compromising the solution's efficiency. The second challenge that our customer is facing is related to the transcription accuracy in both Arabic and Turkish languages. They wanted to reduce the time spent by the translators by imporving the accuracy of the transcription engines focus and provide subtitles that are contextually appropriate for different regions. Addressing these challenges will empower to effectively utilize their existing solution with transcripted files, enhance transcription accuracy for Arabic and Turkish languages, and ensure the high quality of subtitles for their VOD content.

Why Zero&One and AWS

Zero&One's partnership with AWS represents the pioneering of Generative AI (GenAI), combining Zero&One's deep GenAI expertise with AWS's robust cloud capabilities. This collaboration enables the creation of sophisticated GenAI applications that transform data into innovative solutions, driving efficiency and personalization across various sectors. With AWS, Zero&One accesses unparalleled computational resources and AI tools essential for the scalable development and deployment of GenAI models. This collaboration ensures scalable, secure, and efficient GenAI solutions, leveraging AWS's infrastructure for seamless integration and environmental sustainability. The joint expertise of Zero&One and AWS offers organizations a powerful platform to innovate and gain competitive advantage through GenAI, marking a partnership that's leading the way in technological advancement and strategic business growth.

The Solution

Zero&One, in collaboration with AWS, has crafted an innovative solution tailored specifically for our client, utilizing the power of Machine Learning (ML) to revolutionize their data analytics capabilities. At the core of this solution lies the fine-tuning of the state-of-the-art Whisper Model, a cutting-edge pre-trained deep learning model designed for automatic speech recognition tasks. The objective is to elevate the client's data processing and content accessibility by enhancing transcription accuracy up to an impressive 90% for Arabic and Turkish languages. To achieve this, Zero&One's ML experts engage in deployment of fine-tuned the Whisper Model.

By adapting the model to comprehend the unique dialects, accents, and contextual variations present in both languages, Zero&One ensures that the model produces highly precise and culturally relevant multi-language subtitles for our customer's diverse audience. The fine-tuned Whisper Model is seamlessly deployed on AWS SageMaker, providing a scalable and reliable infrastructure that hosts the ML model as an asynchronous endpoint. As an industry-leading machine learning platform, the solution efficiently processes transcripted (.vtt or .srt) files, removing the need for additional audio processing and accelerating the subtitle generation process.

AWS's robust infrastructure ensures consistent performance and reliability, even during peak usage periods, allowing to handle varying workloads with ease. In addition, the solution includes a cutting edge content summary feature powered by Generative AI (GenAI) technology. This feature automatically generates concise summaries of the transcribed content, providing a quick insights and enhancing the accessibility of information.

Solution Architecture

Results and Benefits

The solution brings remarkable improvements to data and analytics for our client. Through meticulous fine-tuning through seamless deployment, Zero&One has tailored the solution to excel in handling the intricacies of Arabic and Turkish languages, achieving an impressive transcription accuracy of up to 90%. By deploying the fine-tuned Whisper Model on AWS SageMaker as an asynchronous endpoint, our client experiences a streamlined workflow for subtitle generation. The solution automates the processing of transcripted (.vtt or .srt) files, eliminating the need for manual transcription and significantly reducing turnaround times. They can now efficiently manage a large volume of content, optimizing resource utilization and meeting tight schedules with ease. With the integration of GenAI for content summarization, our client gains a competitive edge in managing and analyzing their data. This feature not only enhances workflow efficiency but also provides quick and concise insights.