Litigation is a machine which you go into as a pig and come out as a sausage - Ambrose Bierce, allegedly
|
|
|
We test drive Google Voice Search
OPEN SOURCE SMARTPHONE OS Android has received an update to its voice recognition software allowing more actions to be carried out hands free, so we've taken for a spin to see how we like it.
Previously it had been possible for Android to recognise basic voice commands, however the latest update allows Android 2.2 'Froyo' users to issue far more than just search orders. This version supports voice commands not only for calling and web search but also navigation, music playing and dictation of SMS messages and notes.
Curious to find out how well we could bark orders at our phone, we downloaded the application from the Android Market onto a Nexus One and set off trying to bring our phone to heel using voice recognition.
Initially it's hard to see what's different as the voice search application is still represented by the small microphone icon next to the search widget. An understated appearance belies what is actually a very powerful application. Now commands are issued in the form of: command, argument.
For instance, if you wanted to send a SMS message to John Smith, the command would be "send text to John Smith hello John". The commands are pretty intuitive so learning them shouldn't be a problem. There is an issue, though, and it has to do with accurate identification of names, places and words.
Even when in a quiet room, the system is flummoxed by pronunciation of names, either resorting to a search box with possible matches or simply taking an age to analyse the command. In environments where there is background noise, the system is even more hit and miss. Given that the Nexus One has two microphones to offer better noise cancellation abilities, the performance was disappointing.
When it came to navigation, voice commands usually resulted in a list of possibilities being shown. That's not particularly surprising given the repetition in street names, however it leads to the question, if finger input is required, why waste time faffing around with voice commands in the first place?
Indeed, issuing commands by voice seemed to not only prolong our interaction with the Nexus One but increased our frustration at seemingly being unable to carry out the most basic of tasks in that manner.
The software itself is pretty lightweight considering what it does. With the powerful 1GHz chip inside the Nexus One, voice analysis was quick most of the time. Jumping to GPS navigation from the Voice Search application was particularly impressive, with very little lag.
The problem with Voice Search was that many times we found that it would simply be quicker to input the text by hand rather than repeat words at the phone. With a noticeable deterioration in accuracy when in noisy environments, shouting orders at your phone still remains more of a way to show off than useful. µ
Tags: Google
Share this:
Share
If you aren't American, you may improve recognition performance by making a point of saying "a" to rhyme with "hay" and "the" to rhyme with "bee". Otherwise, coughs and ums and scratching yourself tends to be picked up as "a a the a". Added to your text randomly.
Also, I told my computer that I had invented a new word pronounced "perry odd" and spelled "period". The pronunciation "pee ree odd" makes a "." andit didn't seem to be possible to change that.
When peach wreck ignition catches on big, I expect that spoken language will change in many details.
I also concentrated on accurate recognition of cusswords, so that when the thing acts up and you say something to release tension, it doesn't add to it instead. Although you also probably don't want to text those words to a pious aunt. So maybe hook them up to various delete-text commands, if you can. "Chuck" for a single word undo, "chuck that" for sentence, and so on.
Most of this software takes time to learn your speech and contacts/names you use. So how long did you give the application to learn you? Really three or four days maybe to get up to speed. Kind of like Autoweek talking about mileage on a car with less than 10,000 on it.
For the latest version of google voice to work well you need to configure two things:
1) Set the recognition language to US English. Yes, somewhat laughable but true.
2) Ensure that your contacts are all google contacts. The voice recognition system is cloud based and needs access to your contact names - having them as context makes a huge improvement to recognition rates.
2V.
As time goes on, people will probably personalize their computers more and more. So the question is, how many John Connors, Jeeves, and HALs will there be in the near future? Not to mention the Spocks and that head and torso dude I can't remember the name of.
"directions to Sesame Street"
#Did you mean...?
#1 - Suzanne Street
#2 - Susan's Teats
#3 - Southern Trite
#4 - Sesame Street
#5 - Szechuan Treats
"four"
#Did you mean...?
#1 - 4
#2 - 5
"one"
#Oh just use your bloody finger!
I'd expect that - select "1", "2" etc. Maybe Microsoft or someone patented it...