Mobile Search: Talking to the Internet - can voice enabling become the standard way to search?

by: Cian Tuesday, November 18th, 2008

Most people who own a mobile have encountered voice-recognition software at the end of the line at some point. My experience with it has mostly been with paying bills and booking cinema tickets and my experience has usually not been good. Within a few frustrating minutes, I just want to talk to a real person. So are the companies that are investing in voice recognition for their search engines actually producing something worthwhile?

More and more voice-enabled search engines are appearing in the mobile market. Google, Yahoo and Microsoft (through TellMe) have adapted their mobile Internet search engines to accept voice input. Smaller companies like Novauris (Medio) and Nuance offer solutions for companies who want to access voice-enabled search for their services. They all strive towards the same goal. They want their users to ask their search engine a natural question, as if they were speaking to another person, and get a useful response.

Now, I can think of any number of occasions when being able to speak to the Internet would be fantastic. You’re standing at the bus stop under your umbrella in the pouring rain, and you suddenly realise you need to check for a restaurant to eat in that night. You’re keeping one eye on the telly and you don’t want to have to stare at your phone. You’re driving home from work, and you can only use one hand to hold the phone…. only kidding! Even the most nicely designed devices are a bother to enter text into. Mobile phone text entry has gotten a lot better, but whether it’s touch-screen QWERTY or an entire miniature keypad, simply saying your request is going to be a hell of a lot faster.

The problem is that, from what I’ve seen, most of these engines return satisfactory response for maybe 7 out of 10 requests. Now, when I say a satisfactory response, I don’t mean that you find what you’re looking for the first time round. I mean that the search engine correctly interprets your request. Roughly one in three times, the search result you get bears little relations to what you’ve said. This doesn’t necessarily mean that there’s a flaw with the software. It’s just that there are more problems with interference when it comes to voice data entry. There are a lot of words and phrases in English that sound the same but have different spellings and meanings. There are problems with regional dialects and accents. Background noise can seriously mess with voice recognition, as passing engines can drown out or warp entire words and phrases. Text entry doesn’t suffer from those problems.

Ultimately, while it can be a very handy time saving tool, I seriously doubt voice-recognition will ever pose a serious threat to the dominance of text search. Text search is more precise, more direct, and less open to interference error. The most successful implementations of this technology are the ones that combine voice and text, using voice to speed up the process and text to narrow the search.

Of course, if you really do need a search device that 1) understands natural English perfectly,

2) can have search details explained without the use of text and

3) deals with interference more easily, you can always ring your human operator!