Give your voice – different dialects will help train AI

Give your voice – different dialects will help train AI in the future

A new project requires Danish voices to help programmers better understand us in the future.

At Dokk1 in Aarhus, they have set up a booth where you can donate your voice. (Photo: © Emily Holland-Fischer) 5 minutes ago

’I didn’t understand, could you repeat that?’.

Have you heard a robotic voice respond with that, when using voice recognition, it's likely because it hasn't learned the nuances of your dialect.

Voice recognition, such as GPS or digital assistants like Siri or Alexa, can understand and respond to our commands to a greater extent.

However, they may not be well enough trained to understand all those who do not speak standard Danish, as Bo Fristed, chairman of Open Data DK, who is behind the collection of voices, explains. (http://donerdinstemme.dk)

- If you have a slightly different way of speaking, it can be difficult to be understood when calling your bank or insurance company, or in the future when self-servicing with the municipality.

And we are seeing into a future where we will increasingly come into contact with artificial intelligence when we call the public sector.

-----------------------------------

Give your voice for a good cause

-----------------------------------

Open Data DK, is a Danish association of municipalities and regions, collaborating on opening their data in a common open data portal. They are behind the project, which aims to collect 50,000 donations of voices. (https://www.opendata.dk/blog/give-your-voice-with-dialect-accent-and-all-its-nuances)

They do this to train Danish AI algorithms to understand Sønderjysk, Vendelbo dialect, and Funen just as well as standard Danish.

- We are trying to look at this a little like a national collection of voices, where you donate, just as you do in many other contexts when donating to a good cause, says Bo Fristed, chairman of Open Data DK.

When the project is completed, it will all be made freely available, so that in the future, fewer speech-based solutions can be developed.

Facebook (https://www.facebook.com/share.php?u=undefined)

Warning: This article was translated by a Large Language Model, in case of doubt, you can always visit the original source.