Tuesday, October 18, 2016

It's good to talk


Interesting story in the computer press today about a software advance at Microsoft. A team from the Artificial Intelligence and Research department announced that the latest iteration of their speech recognition software had achieved an error rate of 5.9%, i.e. in a segment of normal human speech (i.e. conversation) it interpreted 94% of the words correctly, this is pretty similar (if not superior) to the rates you would expect a professional transcriber to achieve. This achievement, along with some other advances in NLP (natural language processing) and machine learning, opens the door to potentially having a proper spoken dialogue with computers rather than having to "type" stuff all the time. Removing the barrier of the keyboard and operating systems of modern computers opens them up to communities and uses that have proved difficult to satisfy, for example handicapped and elderly people. The race is now properly "on" to achieve fully integrated, voice controlled systems that do real and useful things in the real world. 

I recently got my hands on the new Amazon Echo device which admittedly has a limited set of topics currently, but has speech recognition and natural language processing capabilities that are excellent for the price point IMO; and it gets better as it learns the nuances of accents and FAQ's etc. You can simply ask it things like "how do you spell xxx" or "what's x divided by y" and you instantly get the right answer (a boon for my kids doing English and Maths homework!) I'm also finding it useful to capture shopping list and "to-do" items whilst juggling pots and pans around the kitchen (as the thought strikes!); I also rate highly it's music playing abilities, combined with a service like Spotify pretty much any track from any era and any artist is available just by asking for it. My next experiment is to link it up to some home automation devices, see if I can control some of the more frequently used devices and lights around the house. It's a brave new world, if I were 18 years old again and wondering what field to get into at this point in history, I think AI would be it.

No comments: