46 comments

[ 2.8 ms ] story [ 110 ms ] thread
Combined with the best search engine in the world, this is nothing short of magical. We are living in the future.
Interestingly enough, I notice it on http://www.wunderground.com/ homepage using Chrome.

Not exactly sure how they're using it?

Now all we need is google instant with voice capabilities. I would like for it to predict what I am going to say.
"Please note Voice Search is available for US English only."

Damn, and there I was going to search for Crocodile Dundee quotes.

I can't go to google.com without being automatically redirected to google.co.uk. (I'm in the UK).
> I can't go to google.com without being automatically redirected to google.co.uk. (I'm in the UK).

If you replace the 'co.uk' with '.com' after you've done a search, then it should stick.

There's a link at the bottom which says Go to google.com.
Thanks - link at the bottom (I'm blind!).
I can't believe people don't know about google.com/ncr, which disables country redirect
Just went 0 from 4, anyone getting good results? Not ruling out a bad mic or pronunciation on my end, although I imagine because searches are short and abstract they could be pretty hard to parse.
I sent a bunch of Cocoa classes at it, for no other reason than that's what's on my mind right now.

NSView -> "nfcu". OK, fair enough.

NSViewController -> "nfcu controller". OK, I guess.

CAAnimation -> "salvation".

This is probably a place where that controversial "profiling" could probably help, since I'm an atheist Cocoa developer.

My first test was "Look at my horse my horse is amazing" and it worked just fine even though I am not a native english speaker.
I said, 'I cant be cool,' but it came out 'I cant b cool'. Extremely close in my opinion, and Prince would be proud. For more generic words the results came out perfect: 'wired', 'baseball', 'hacker news', and 'Philadelphia Phillies'.
Gonna be great for searching lyrics
I noticed I finish searching by typing faster and more accurately than speaking into the mic. I understand the data collected by Google is valuable for its own use but what's the real benefit to the user?
Theoretically they use said data to make the speech search better.
Too bad it' hardly working though. =( 7/10 was totally failing.
Didn't work for any query. Crashed my browser once (Chrome 14.0.794.0 dev / OS X 10.6.7).
Works just fine on 14.0.797.0.
Yeah, I got "Hacker News" after four times. Now it would be nice if I could say "Open" then it clicked on the first result.
This is a big step – millions of people becoming more comfortable with searching for information just by speaking to their computer. Historical moment.
This technology has been around a few years. If I'm not mistakken, wasn't voice search originally implemented on the Google iPhone application? Now with the proliferation of microphones on computers, it makes sense for Google to implement this feature on the Google webpage.
Yes, technology was around for years but it was just made more accessible to a much larger demographic.
Does anybody know a keyboard shortcut to activate the voice search without having to click on the microphone?
Yeah, seriously. Also, from the Google web site: "Searching without typing Use it in the kitchen, in the garage or anytime your hands might be full."

Oh, really? I want to see you move your mouse pointer into position and click the microphone...anytime your hands might be full. Silly bints.

(comment deleted)
Did anyone else feel uncomfortable with Chrome turning on the mic, recording sound and posting it to a foreign server without asking whether it was allowed to? Feels like this could be abused. Would have appreciated an "Allow goolgle.com to access your microphone?" security question.
I'm ok with it. Much in the same way I don't expect to be asked if it's ok to send data to their servers every time I do a normal text search.

In what way do you think it could be abused?

It's one of those web things, without a prompt it feels like any site ever could just recording through the mic and saving it.
Didn't you have to click the mic icon before it started recording anything?
Yes, but I don't know what the mic icon is. Does it just call a function? Is it a Chrome-provided icon that is always on top? Can the menu that is shown be edited by the website?
It calls a browser function, not a javascript function.
I don't think it allows any arbitrary website to record your sounds. Your browser is in charge of turning your sounds into text right?
without asking whether it was allowed to

I'd hope the mic icon and popup are part of chrome, rather than being something implemented by the web page. If that's the case then consent is implied by clicking on the mic icon.

Well apart from "coca cola" it didn't get any of my search phrases right, a couple of words that seemed to be based on past search history though ....

"paulstretch" = balls "poor" = p* [presumably 'porn'?] "text to speech" = x

etc.

Just tried "pepsi" and get 'speech not recognised'. But for "pepsi cola" I get the right search ...

Non-native English speaker:

  Haskell web frameworks - √.
  Tigers in India - √.
  Friedrich Nietzsche - what is a meter.
  Programming best practices - √.
  Contemporary philosophy - √.
  I bet you look good on the dance floor - √.
  Random beatboxing - the gap at the top of the pita pit.
Native English speaker – I had the same problems with non-Anglo names. It seems to discount the possibility that one is searching for a foreign name (in my case, an Italian name), and instead tries to interpret the name as if it were an English phrase.
On blind faith alone, I have to imagine this will be improved significantly going forward.

I use Google's voice recognition algorithms/systems on my Android phone all the time. Every day. Often to dictate complete SMS and Emails, while I'm driving, via my car's AD2P bluetooth mic, with all the ambient noise/etc present in a car on the interstate. (Though to be fair it's not an aftermarket bluetooth solution and I'm sure the mic is high quality and well placed.)

It is very good. The most common type of mistake -- and still rare at that -- is stuff like "let's not go until..." is transcribed as "let's not going tell" and stuff like that.

When I first started using the voice features when I got the phone (HTC Evo), i was blown away at the quality of the results.

And of course, Google has been practicing this stuff for a very long time. Everybody knows about Goog-411 (which was heaven before I got a smartphone) but does anybody else remember the little gimmick Google Labs page where you can click on the page, then call the number it displays for you, speak your query into the phone, and then right before your eyes your page reloads with your query and results.

That was at least nine years ago.

It works pretty well for calculations. Tried phrases like "natural log of 77", "80 miles in kilometers", "17 times pi" and got the correct answer. Cool!