My Accent is an AI voice recognition system with a difference; instead of trying to ignore the differences between voices, it looks for the differences between them. In this case our AI was trained to differentiate between British and American accents. The user speaks a phrase and it shows which accent it thinks you have and its confidence in the decision. It is impressive, even though only a few hundred speakers were used for its training. Since My Accent's launch in January, over a million accents have been classified.
Can you fake it?
Needless to say, during testing people tried their best fake accents, and found it harder to fool the system than they thought!
Explaining the result
Once the system has told you whether it thinks you're British or American, it goes on to explain why. This is a peek inside what is often termed the AI ‘Black Box’. You can see how the decision was made – your text is shown on the screen, and it is colour coded to show which words influenced the system's decision towards American, which towards British, and how strongly.
Why is this important?
The spoken word is rich with nuances that we take for granted. Humans discern age, origins and mood without even thinking about it. Deeper understanding of speech, not just recognition, opens the door to better voice interfaces.
The second aspect of the demonstration is Explainable AI (XAI), making the ‘Black Box’ decision transparent. Brits may be familiar with a series of comedy sketches whose punch line was "computer says no" – never to be explained. Explainable AI will tell you why the computer said no. It gives insights into how the AI functions, providing information to help debug the system when something goes wrong, and eventually could become a tool used by regulators and testers in evaluating the AI.
Footnote: When I speak normally I'm classified 96% British, and my best US accent? Still 72% British.