Posted
Apr 29, 2008
 | By
Andrew Cantle*

The future of mobile speech recognition

April 3, 1973 is considered to be the birth date of the mobile phone. This was the moment when Martin Cooper, a Motorola researcher and executive, made the first call on a heavy portable handset. Fast forward 35 years and look where we are now. Approximately 94% of Australians own a mobile phone but, more importantly, the number of services available on a handset has skyrocketed. We can now access the internet and TV through our mobile phones and a mobile phone can even double as a credit card!

The past year has seen another big step forward in mobile technology with the introduction of voice-to-text services created by SpinVox. SpinVox Voice-to-Text was the brainchild of Christina Domecq, CEO and Daniel Doulton, chief strategy officer. The idea was born when Christina had a moment of inspiration after she emerged from a meeting and found 14 voicemails on her phone and thought to herself: Why can’ t I get my voicemails delivered as text?

SpinVox is way more than just another speech recognition product, the sort you might use on your PC for word processing. SpinVox’s Voice Message Conversion System (VMCS) automatically converts voicemails delivered in English, Spanish, French or German into text messages so the user doesn’t need to waste time listening to voicemail messages. In an instant they can quickly see who has called and view the messages as an SMS text message or as email on their PC or Blackberry.

To be honest, the wow factor comes when trialling SpinVox. I tried it for the first time during my first job interview with SpinVox and I was instantly converted.

I have worked in this industry for over eight years in both the UK and Australia and have seen a lot of new products and services launched in that time; however, SpinVox is a real change in the market. I am a frequent flyer, travelling back to headquarters in the UK, to New Zealand and across Asia–Pacific, and the service proves itself every time I travel. As soon as the plane lands I receive all of the messages left by colleagues, customers and family within seconds as text messages, without having to listen to my voicemails and frantically make notes of who called all while trying to disembark from the plane!

We affectionately refer to the VMCS system as D2 (The Brain). It is a combination of artificial intelligence, voice recognition and natural linguistics technology. D2 can also learn from humans. For example, if D2 doesn t recognise a certain sound — perhaps because of the person’s accent or the fact that it is actually a new word — it asks for a helping hand from a human to identify the sound so the system learns and does not have to ask again.

SpinVox has a dedicated automated speech recognition (ASR) team constantly updating its dictionaries — adding words to the system in every country where the technology is available. In Australia, the system has had to learn a lot of 'Aussieisms' — words and phrases that most Australians wouldn’t think twice about and use commonly in everyday speech.

Physical stereotypes such as Crocodile Dundee, Sir Les Patterson and Alf from Home & Away may have almost disappeared but their language hasn’t. Words like "strewth", "you beaut" and "crikey" are still everyday parlance for Aussie users of SpinVox. And based on the evidence of D2’s conversion, these words are here to stay!

So, as much as Aussies hate to admit to this particular stereotype, words like "grouse", "fair dinkum", "sheila" and "g'day", are being added to D2 . What’s also interesting is that the English aren’t the only ones to use rhyming slang. Australians seems to use it just as much, with terms including:

  • Harold Holt (bolt)
  • Hit the frog and toad (hit the road)
  • Reg Grundys (undies)

Since the launch, the system has added hundreds of Australian terms, including:

  • Bluey
  • Budgie smugglers
  • Chuck a sickie
  • Esky
  • Icy pole
  • Stubby
  • Trackie daks
  • Bottle-o
  • Bogan

It was also fascinating to see the number of words that Australians shorten, including:

  • Brizzie
  • Cozzie
  • Oldies
  • Pokies
  • Rellies

Before launching in Australia, the SpinVox ASR team also added the names of countless Australian suburbs, railway stations, banks, sports teams, and even Australian celebrities.

SpinVox users currently report a 97% accuracy rate in converting voice to text and the company is working hard to further scale and increase the accuracy of the system. Dr Tony Robinson, who has been working at the UK’s Cambridge University in the field of speech recognition since 1985, recently joined SpinVox as director of the Advanced Speech Group. Part of the brief for the Advanced Speech Group is to continually evolve the SpinVox system so that it can understand new accents and dialects. I try to talk to Tony whenever I am back at UK headquarters as it's always fascinating to see what words D2 has picked up.

Australia is a linguistics mixing pot with English and American influences. English words still predominate — think road rather than freeway and mobile phone rather than cell phone; however, the influence of mass media and the increasing influence of American TV and music is changing Australian pop culture and American words are creeping into the Australian vernacular.

SpinVox will soon be entering the New Zealand market and I'm sure there will be lots of interesting new words for D2 to learn, including a few of my personal favourites: chilly bin, jandals, togs, flating and we can’t forget fush and chups!

*Andrew Cantle is the general manager ASIAPAC, SpinVox