COMMENT
"Obviously a doctor who uses more unique terminology and a more expansive dictionary will require more human intervention than the kind of average Joe on the street," Domecq told us at the demo - a qualifying appendix to her assertion that fewer "human interventions" are required the longer SpinVox has been in a market.
Maybe fewer then, depending on the type of professionals who are signing up to the service and the things they are discussing.
The phrase "it varies" came up at a lot in conversation at the demo.
What proportion of messages are fully automated? What proportion need a human touch? What proportion are totally unintelligible and not touched by machine or human? It varies and it depends, we were told time and again - with variables including product type, market, language, time in market, 'newness' of SpinVox user etc etc.
SpinVox was not able to show off the system's learning capability as the demo platform was a shrunken splinter of its real-world system set aside for testing purposes and therefore disconnected from all the data the real Brain presumably accrues about its users. So demo or no demo, people will still have to take the company's word for it that its technology gets better over time.
But what of the demo then? How did SpinVox perform?
Read this
Out of a total of four voicemails left in the quiet conditions of the meeting room only the shortest and most basic message (left by Wheatley) passed through entirely unaided - "Hi Rob, can you give me a call back when you get this? Thanks. Bye". It would be a pretty dumb mechanical eardrum that fell at that hurdle in test conditions.
The other three short voicemails all required the assistance of Ellie, an unfailingly enthusiastic human agent who was in the room to apply her dextrous ears and fingers to Tenzing - SpinVox's call centre software where machine predictions go for help.
The first message that ended up with Ellie contained the word 'SpinVox' which seemed to have foxed The Brain. Another was either spoken too quickly or rejected because of the word 'Tesco'. A third drove the system to distraction - admittedly it was a voicemail purposefully encoded in a Texan drawl so see-sawingly folksy it had most people in the room scratching their heads.
But still: the serious point is that accents vary as much, perhaps even more than, vocabulary - so welcome to real world conditions.
The upshot of the day is I came away enormously impressed with Ellie's enthusiasm, her lightning quick fingers, sensitive ears and very human brain. But far less impressed with what a software Brain can hear, perceive and intuit.
One thing's for sure: technology has an awful long way to go to get anywhere near the hardware inside our heads.
But considering the human brain is the product of millions of years of evolution - and not wanting to be too hard on the fledgling upstarts - let's not forget SpinVox's six-year-old Brain is not even a toddler beside it.







Comments
There are 3 comments. Join the discussion
1. anonymous
Reminds me of a comment about stylus writing recognition on a palm I think.
The person basically ended up writing for the software so it was recognised.
Maybe the same will happen over time with all the automated voice recognition.
Not necessarily a bad thing if the software recognition is reasonable. As it might improve the standard of spoken language and normalise it a bit.
2. Jonathan Present
I should imagine that this company has Intellectual Property which they should be happy to reference, such as Patents, and Algorithms. Do they have employees with the requisite expertise to create a system of the sophistication that they claim?
3. Nick Barnett
Excellent article Natasha. Intelligent, dry style, with just the right amount of objectivity. Leaves me with the same conclusion you drew, about the 'real-world-readiness' of SpinVox. Trailing in the wake of it all is the fickleness - or apathy? - of the millions of average joes (and janes) who are now no longer getting their knickers in a twist about Ellie and others eavesdropping on their private conversations.
Ho-hum.