When text-to-speech goes down the drain • The Register

Bork!Bork!Bork! UK water company Severn Trent learned an unfortunate lesson about text-to-speech systems when a robocall to customers went hilariously wrong.

A Reg reader received an automated call warning of potential water discoloration during planned works from January 19-25. The message advised running taps for twenty minutes if the water appeared discolored – standard stuff, if a bit robotic.

In the recording forwarded to us, a female voice told our reader what to expect. All good, if a little robotic.

However, things went off the rails a bit when the robot attempted to read out the URL for Severn Trent: http://www.stwater.co.uk/discolouration.

For anyone concerned, the URL redirects to a secure connection.

The robot took a run at it, but the result was unlikely to be what the water company wanted. We imagine Severn Trent was hoping for “S T Water,” but instead got “S Twatter.”

Text-to-speech blunders are nothing new. This writer has fond memories of home computers with primitive speech synthesizers producing similar gaffes to the horror of listening parents, and later validating Interactive Voice Response Systems (IVRS) to spare customers confusion and embarrassment.

Decades later, and in an era where Artificial Intelligence should be navigate linguistic landmines, it’s almost reassuring that blindly feeding text into a system without human verification can still produce unintended results.

Unless Severn Trent, a UK water company, is about to undergo a rebrand, and the robot has inadvertently blabbed. Calling yourself S Twatter, however, would be a very brave move, particularly in the UK.

Someone should probably add a quality check to that robocall system. ®