Accents in Voice Interaction
We debated for some time back in 2012 before deciding that we would ship the Ubi worldwide. Shipping world wide on a first time hardware product with new cutting edge technology, what could go wrong?
Besides all of the difficulties in the actual shipping (think $100 for a small package between Canada and Australia), some of the major challenges happened when English speakers started to use the device.
We quickly learned that pronunciation, even of the wake up word, varied widely and that it would make sense to allow some customization. Fortunately, one of the benefits of us using an Android device for the Ubi was that all of Google’s speech recognition libraries were built into the device. One of the features that was a big saviour was that we could switch English Speech-to-Text settings to different accents.
It took some time, but we were eventually able to extend this as a setting. Indian English, UK English, Australian English, and other accents were accessible on the device. Google has since improved even more since then and can now accept English from a wide set of users with different accents, even before switching from US to other localities.
Google had also started to be able to detect accent a few years back. It’ll be interesting to see if this data is used to vary the results of searches. It’ll be great to see Amazon and others follow suit when their devices become available outside of the US.