The installation will use Nvidia hardware to complete its aim of understanding every voice, regardless of demographic, accent, or dialect, with Megaport’s hyperscale connectivity at the Harlow campus, just north of London.
Kao Data CEO Lee Myall (pictured) said: “The opportunity to support Speechmatics in this crucial phase of the company’s expansion is an exciting prospect for our organisation, and further underpins Kao Data’s position as the UK’s pre-eminent provider of sustainable data centres for HPC [high-performance computing] and AI [artificial intelligence].”
The supercomputer will allow Speechmatics to expand its neural network research and development, supporting increasing customer demand for its leading speech recognition technology, said Kao Data, which is built on the site of the former research lab that saw the invention of pulse-code modulation and optical fibre telecoms.
Tony Robinson, founder of Speechmatics in 2006, pioneered the approach of applying neural networks to speech recognition back in the 1980s. The company was among the first developers to utilise the technology to process large datasets at volume once higher performance computing appeared in the early 2000s.
Speechmatics’ high-density supercomputer in Harlow will benefit from bespoke colocation, a power usage efficiency (PUE) of 1.2, and it will be powered by 100% renewable energy.
The company’s VP of Machine Learning, Will Williams, said: “Powering modern machine learning research and development requires an advanced computing infrastructure which only becomes more demanding as you scale.”
He added: “As we continue to develop our technological edge through the use of self-supervised learning in our models, it’s crucial to ensure our data centre provider can scale with our compute demands, but in a sustainable way.”
