It’s rare these days for companies not to rely on data in one way or another. Yet for companies building AI services and applications, particularly those which sit at the very cutting-edge of this sector, their fundamental reliance on data can be both a blessing and a curse.
It can be a blessing in the sense that at no point in history has data been in such abundance, and been so powerful when used to train algorithms. It’s a curse in the sense that this abundant data needs to be stored, moved, managed and secured in ways that pushes the limits of even the giant cloud providers, something Aaron Jones, the cofounder and CEO of Yepic AI knows all too well.
Yepic AI is a pioneer in text-to-speech, or more specifically text-to-video algorithms. After starting as a research project at Founders Factory, Yepic AI became the first of its kind to be spun out from the accelerator.
The power of Yepic AI’s algorithms and machine learning models lies in allowing businesses to turn text scripts into professional videos. In these videos, actors appear to be reading uploaded scripts as-live. This whole process can take mere minutes, with no need for studios, camera crews or production days. Use cases range from introduction videos for customers, to training courses for staff, chatbots, explainers and more.
More recently, Aaron’s company took this a step further with the release of a product called VidVoice – the world’s first and only simultaneous video translation service. Using the same advancements seen with Yepic AI combined with cutting-edge translation and error-rate analysis, VidVoice lets people understand and speak any language to one another without interpreters, distracting subtitles or robotic voice overs.
“Businesses tell us that they’ve been crying out for a service like Yepic AI for years, and our recent growth is testament to this,” says Aaron. “We’ve seen similar results from early trials with VidVoice, and we hope this will continue when we open the public beta towards the end of the year.
“However, the true scope of this technology goes beyond helping companies. 25M people in the US don’t speak English, and their healthcare suffers because of it. Instead of having to rely on friends or interpreters, our technology would allow them to have native conversations with anyone in the world. This is one of many benefits we see stemming from our advancements and is why we’re so excited about it.”
The power of data
In many ways, what Yepic AI is doing is making the kinds of services seen in sci-fi a reality. Yet what sci-fi series and films fail to highlight is just how data and power intensive such breakthroughs are.
The amount of data needed to process video generally is vast, let alone when this is run alongside advanced machine learning algorithms. This, in turn, means such processes require almost unprecedented levels of computer power, storage, memory and support: the kinds of levels provided by hardware and software services that far exceed the budget of a growing scale-up.
To navigate this, Aaron recently partnered with OVHcloud. OVHcloud is a European cloud provider with more than 400,000 servers housed within over 30 data centres across four continents. By taking an integrated approach, OVHcloud has supported 1.6M customers in more than 140 countries.
Although best known for its Infrastructure-as-a-Service (IaaS) offering, the company is seeing increased interest for its Platform-as-a-Service (PaaS), particularly from startups specialising in AI. This is because it combines the best of both worlds when it comes to control and access.
Developers and programmers get a platform designed to help them with their daily work, without having to maintain the infrastructure. That’s all handled by OVHcloud. Not only does this give instant access to notebooks, GPUs and scalable storage, but it also means businesses don’t need to invest in local servers or train maintenance teams in order to use their tools.
It also allows developers to concentrate on what matters most – building the services and applications that make a real difference to customers, and wider society.
For startups, particularly, where a trusted cloud is key to growth and the costs of such services can be beyond the reach of early-stage firms, OVHcloud offers the OVHcloud Startup Program. This programme gives founders credits over a 12-month period to enable them to experience the full power of its products. Since its launch in 2015, the programme has now hosted over 2,000 startups, selected from more than 7,000 applicants. This includes Yepic AI which was granted credits of €100K from OVHcloud as part of the programme.
To get started with OVHcloud’s products, users like Yepic AI log in to the platform via the OVHcloud Control Panel, or an API. They can select which services they use, and which they don’t need. They’re only billed for what they use, and they get personalised support. Having been stung by spiralling costs and a poor experience with the cloud giants, it was OVHcloud’s transparent nature that attracted Aaron and Yepic AI.
“Cloud giants provide credits for startups but when you’re running machine learning models, you soon rinse through those credits,” says Aaron. “With OVHcloud, we only pay for what we need, and we can monitor this usage easily. This is why we trust OVHcloud with all of our machine learning and generating video off OVHcloud.”
OVHcloud’s AI Notebooks allow Aaron’s team to rapidly test and deploy at a small scale.
“We can push unstable things live and test them in real-world scenarios before refactoring the code and repeating the process,” he adds. OVHcloud then integrates GPUs into the wider solution, to give Jones’ team on-demand access, and these GPUs have been adapted to suit parallel processing, which is a major requirement of machine learning and deep learning.
OVHcloud handles every stage of Yepic AI’s data lifecycle, from storage to transfer, intensive use and deletion, while guaranteeing confidentiality, sovereignty and security. This is already needed, but should VidVoice be used in healthcare settings, the protection of private data will become a fundamental necessity.
“For companies of all sizes, MLOps and DevOps are expensive and difficult to get right, yet for startups and scaleups, this can be the difference between thriving or surviving,” says Aaron. “OVHcloud’s PaaS offering helps us maximise our internal resources, freeing up our researchers and allowing them to test their code and push it live at rates that wouldn’t be possible without it.”
You can learn more about the OVHcloud Startup Program and apply here. We are also running an AI focused showcase event on 10 March which offers expert panels, startup and scaleup pitches, networking and access to marketing and funding. Sign up here.