With the increasing use of analytics and voice-enabled solutions, a whole new market for speech-to-text technology has opened up. To make sure it didn’t miss out on valuable business opportunities, Cedat85 decided to speed up the learning process of the neural networks underlying its technology, which used highly complex algorithms. Cedat85 opted to use IBM’s AC922 Power Systems, which are designed to provide high-level performance for artificial intelligence workloads. After running a test on these servers, Cedat85 found that they significantly improved the working environment by increasing the process speed by 2.4x within a matter of days the platform was installed.
The reduction in training time also reduced Cedat85’s risk margin, as engineers no longer have to wait long to figure out a neural network needs a tuning, and the company was able to take advantage of new opportunities.
‘IBM’s AC922 Power Systems servers have allowed us to experiment more, we can try different things without worrying about setting up the infrastructure in such a way that it can dedicate itself to training an imperfect model for a whole month.‘ said Enrico Giannotti, CEO of Cedat85. ‘We can also be more proactive with our customers: this helps us to grow and get a return on investment within a few months.‘
Afterwards, Cedat85 was awarded a contract for the British Library’s “Save our Sounds” project, one of the most important and complex projects handled by the company. The British Library’s broadcast archiving process required models in use to be enhanced, as they needed to recognise different inflections, dialects, and colloquial terms that exist in British English. Thanks to the collaboration with IBM this was made possible considerably faster.