Listen to this story
Synthetic data is used by several organisations today. The advancement in signal processing and machine learning algorithms have resulted in high-quality synthetic data generation techniques. One such deep tech start-up is ‘Kroop AI’, which has revolutionised audio-visual synthetic data for businesses through AI.
KroopAI helps businesses and organisations generate audio-visual content using an easy-to-access, cloud-based studio and API. The company utilises a synthetic audio-visual deep learning platform to generate high quality videos with solely audio or text as inputs.
Subscribe to our Newsletter
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
In addition, it helps in the creation of avatars with unique audio-visual generation on Metaverse, also providing accurate lip-sync while dubbing content. KroopAI’s state-of-the-art ethical audio-visual data generation technology, combined with multimodal and explainable deepfakes detection, also protects client authentication and verification systems.
In February 2021, Dr Joshi established ‘KroopAI’ along with co-founders Sarthak Gupta and Milan Chaudhari. Since then, the startup has carved a niche for itself. KroopAI was one of the five Indian startups to pitch at ‘Cannes Next’—an executive conference exploring the future of the entertainment sector, held in May 2022.
Dr Joshi says, “KroopAI enables organisations to generate audiovisual content in 2D or 3D formats through text evidence. Organisations can easily bypass the requirements of having a studio setup or making purchases, with our trained inline model that can be used anywhere.”
A PhD from the University of Canberra, Australia, Dr Jyoti Joshi is a first-generation entrepreneur. She has worked extensively for mental health analysis and has many impressive experiences from University Waterloo, Canada and Monash University, Australia.
Dr Joshi emphasises, along with KroopAI’s co-founders, that her primary goal is to look after the overall strategy of the company. “Based on the discussion that we have on the product side, we discuss the new things that we can work on—such as different sectors and verticals.” She further explains that her main goal is to envision how KroopAI might evolve, not just in the near future but also in the many years to come.
The firm’s audio-visual deep learning based platform, ‘The Artiste’, generates high-quality videos with digital avatars using audio or text as input. These generated videos gear the platform
towards creating data for the MetaVerse. Through its VizMantiz API, the platform also enables organisations to detect unethically generated synthetic media.
How was KroopAI launched?
In 2017, researchers from University of Washington presented an article at SIGGRAPH 2017 that produced photorealistic results to demonstrate the synthesisation of a target video clip to match a given audio track and thereby altering the original audio-visual clip to match the said audio track. The result shows how AI was used to match the given audio to generate a clip of the former US president Barack Obama, complete with realistic mouth shapes while talking.
Dr Joshi describes this as the eureka moment on her journey to launch ‘KroopAI’.
“Having been in academia for over a decade, I did contemplate moving into the industry. The pandemic gave me a lot of time to think about doing something more impactful. The whole idea was to really use my research to bring out something that can be used with a cause. And that was the time when we came across Barack Obama’s video where we realised that a lot can be accomplished in this space.”
Dr Joshi explains that there are various ways to leverage technology, with the generation of synthetic materials posing several challenges. It could be as simple as meeting the requirements of demands for the integration. “We want to see more and more new content that’s exciting, with a need to have personalised endpoints. That is the technology that we use.”
Along with the challenges of her field, Dr Joshi also addresses the challenges that entail in her role as the CEO—“It’s hard to have a positive environment around all the time. When you are building something from scratch, one would have certain expectations set. Making use of technology is one thing—but how do I package it and sell it? More than the solution, I am also selling the technology.”
Dr Joshi believes in using her research skills to speak about the company’s challenges and how technology is the solution for it. “Sometimes things don’t go the way as expected, but with better time management and understanding within the team, [it usually] works out.”
Further addressing the larger discourse about the good and bad aspects of technology as an AI scientist, she remarks that, “It’s an open debate. There could be different ways this technology can be exploited in the financial sector with frauds. Another one would be the hacks that happen in the software or the hardware industry. There could be a fake video moving around—a synthetic generated content, and that’s how people form their opinions that could motivate people to take action. This alters people’s projections from the real from synthetic generated content.”
AI research journey
To formally learn the nuances of pursuing research, Dr Jyoti Joshi pursued her PhD from the University of Canberra on ‘Multimodal Assistive Technologies for Depression Diagnosis and Monitoring’, wherein she worked on creating tools that could be used as objective measures of depression. “It has been a fascinating 4 years to work on this particular project.”
Dr Joshi started to apply artificial intelligence and machine learning specifically to depression analysis. “The challenge that I faced was the nature of data that I had in my hands, with large content to review that pertained to a very small population.”
While pursuing her degree, she also visited the Queen Mary University of London and the University of Pittsburgh to experience different AI-centric research environments.
Reminiscing these experiences, Dr Joshi said—“I realised that knowing your data is really important, whatever AI ML problem that you solve, one should know how complex your data is. If there are larger data or datasets with challenging features, one should figure out what sort of learning algorithm has to be used.”She also believes that deciding the appropriateness of the right learning algorithm is crucial, since not every algorithm is appropriate for each use.
“You could be using very fancy things for simple tasks to be done. With whatever current problems that you are tackling, there are different ways you can solve them. Knowing your literature is also equally very important.”
On gender inequality
Dr Joshi talks about how there’s an urgent need for an overall representation of women in the analytics space, and across fields of engineering, physics, and science. She believes that organisations must focus keenly on assisting women who pursue their master’s or research degrees to excel in their respective fields.
“Every organisation wants to have equal representation. In my opinion, it’s the mindset where society has to understand that we should not be creating equality, but equality should exist.”
Dr Joshi recommends that organisations should be equipped with mentoring programmes that involve good participation by all individuals, with both men and women equally being part of the experience. She believes that the industry might benefit greatly by tapping and attracting those individuals who consistently excel in the AI/ML field.