Here is another fine irony of Lithuania’s digital transformation. Lithuania has a digital fibre-optic highway — yet data from the LIEPA-3 project, which cost more than €4 million, enters it through the garden gate of Synology QuickConnect, pushed along in a wheelbarrow at roughly 0.91 MB/s. We can have a gigabit connection, or even ten gigabits, but none of that makes a difference if, at the other end, a 1.13 TB national speech corpus is being shared through infrastructure better suited to a folder of family photos than an AI dataset of national importance. The project’s promotional slogan declares: “Teach technology to speak Lithuanian!” At this rate, technology will learn Lithuanian on its own before we finish downloading its training material.
A catchy slogan does not guarantee technical accessibility. Training AI requires automated, reliable and reproducible data retrieval — direct URLs, checksums, and resumable, parallel downloads. Not a browser window, a password and two weeks of hoping the connection will hold. With a budget of more than €4 million, could a proper file server perhaps have made it into the funding plan? So yes, Lithuania’s internet is fast. Apparently, nobody has told LIEPA-3.
Link mentioned in the article: Development of a Lithuanian speech corpus (for speech recognition)