Dubverse is a leading media company with over 120,000 users across the world. They provide a range of GenAI solutions for users to create content, with one of their premiere products allowing users to transcribe any video with great speed and accuracy. 

As they continued to grow, their existing third-party transcription partner was unable to match their increased workloads, and the cost was starting to become prohibitive.

When working with transcription use cases, in addition to performance and quality concerns due to generic solutions, data privacy is also a worry. Calling external APIs to handle user data can expose a company to risks of data breach and confidentiality. These would be mitigated by a performant, cost-effective, on-prem deployment.

By relying on an API provider for transcription, Dubverse faced some key challenges:

• Scaling transcription costs that were starting to make unit economics unfavorable and taking away resources for product development and growth
• The API provider was unable to maintain the speeds required as the number of requests increased, and with data leaving Dubverse’s cloud, there were data compliance issues.

Simplismart conducted an in-depth analysis of Dubverse’s load profile and use case to transform their speech-to-text workflow with Simpliscribe. Simpliscribe uses a proprietary inference engine to deliver more performant deployments of popular open source models.

• An optimised deployment of OpenAI’s Whisper allowed for a more performant workflow, increasting transcription speed by 36%, and reducing costs by 90%
• Simpliscribe is finetuned to look at the phonetics of the input speech rather than just the words, and this coupled with training on lower quality and Indic languages makes the text output much better than off-the-shelf transcription solutions.  
• With a fully on-prem deployment, no user data ever leaves the Dubverse cloud, addressing any potential data privacy or security concerns.