Comment by jpxxx | Hacker News Reader

Comment by jpxxx | Hacker News Reader

This is Next Level Shit. This is absolutely next level execution. The responsiveness is incredible and it immediately falls through to a well-formatted search result if it can't give you a soundbyte or a Knowledge Graph result.

Unit conversions provide in-line converter widgets... it'll gleefully show you pictures of anything safe-search while playing dumb if you search for something "naughty"... web links you select pile up in little tabs that let you slide right back to the original query... it looks good... it makes pleasing sounds that let you know what's happening...

If Siri can stage a question to Wolfram Alpha, the result is great. But if she can't, she just lamely offers a button to (Search the web for ______?) that then kicks you out to Safari. Google voice search makes Siri feel clunky.

The voice recognition is verging on instantaneous. This is amazing work.

> The voice recognition is verging on instantaneous.

Exactly! Shows you how much it has recognized as you speak. Immediate feedback. More important is the speed of getting back results. Google Now is certainly a little faster in that regard as well.

Reminded me of a 21 questions test with Siri and Google Now (on a Nexus with Android 4.1 I think) side by side as they listened simultaneously. It's interesting to see how both perform in different scenarios. Generally when Google search has the answer Google Now is a little faster. When Google search doesn't have a direct answer Google Now boils down to Google Search results - right there - as opposed to an offer to do a web search. In my experience Siri tends to have an answer more often.

The general theme was:

1. Questions that get answered by Wolfram Alpha were slower on Siri. In one test Wolfram Alpha didn't have an answer and Siri offered a web search where as Google's graph did have an answer.

2. For requests like "Call BestBuy" Siri looked up BestBuy stores and offered you the choice of which one to call where as Google Now said there were no numbers for BestBuy in the phonebook.

3. Sports questions (which I don't quite care about personally, sorry) tend to be answered way better by Siri. I remember a question about which sports personalities was taller and Siri had an direct answer with other stats whereas Google Now was a list of search result links - no ads yet. It was the same thing for another sports trivia question.

4. Siri would sometimes have to wait - almost like a timeout - on a response from Apple. Google Now didn't face such issues.

5. In some case Siri would take a little longer but give back more information - when you talk about restaurants it'll include reviews, price range and distance.

Source: http://youtu.be/z_pclCFpjgw (video)

>> "Exactly! Shows you how much it has recognized as you speak."

That's the one thing about it I didn't like. The fact that I could see a mistake made me want to try and correct it (but I don't see any obvious way to say 'that last word is incorrect').

If it acts like Android voice recognition, it is smart enough to correct speech on the fly.

There is something about the voice recognition algorithm that uses later words to understand what you were saying earlier. For example, whenever I say, "Show me pictures of humpback whales" - GoogleVoice starts off with "Show me pictures of home" and then when I add the word "back", it corrects home to be hump.

To some degree, that's probably how the human brain works. If you say out loud "Show me pictures of Hump" and then just stop, it really does sound pretty close to "Show me pictures of home"

I am very, very impressed on how much better than Siri the voice recognition is in terms of speed and responsiveness. And, it works just fine on my iPhone 4, where Siri isn't an option.

The standard machine learning approach for doing exactly what you describe is to use a conditional random field. I don't think wikipedia has a great page on it , but check out http://en.wikipedia.org/wiki/Conditional_random_field for a jumping place.

CRFs are used in places like voice and images where recognition or decoding of a segment logically depends on the pieces near it.

For clarification, it's much more likely a Hidden Markov Model. Most speech recognition algorithms use some form of Baysesain probability model, HMM being the most commonly used e.g. Sphinx CMU.

Wow. I haven't tested IOS6 with SIRI, but when it was Google Voice v/s Siri on IOS 5.1.1 - there was no comparison. Google voice won hands down. Even in a noisy environment. I've just bought a second hand iPhone 4s that will be upgraded to IOS6 so I can compare for myself. Until a month ago, Google was easily king of the hill.

Remember Google has no incentive to get you to upgrade your phone, Apple does.

Yes, but Google has incentives to display ads. Which is the lesser of 2 evils?

The former. I want to see relevant ads.

Its pretty easy to ignore ads...

Agreed. I'm running an iPhone 4. The speech recognition almost keeps up to my speech in real time. The results are typical Google.

This application gave me one of those "sufficiently advanced technology" moments. Magic.

agreed. I want this to be the triple-click function on my iPhone 4.

Indeed. I really wish there was a way to use this instead of Siri with the standard home button shortcut. Pretty much all it's missing is a way to do basic phone interaction (call X, text Y, remind me tomorrow) which should be pretty simple considering they apparently have real time voice recognition.

It's not simple on iOS since iOS simply does not allow it.

You can get pretty close, the tel URL scheme will let you make calls. MFMessageComposeViewController will let you send texts. Calendar stuff is already possible with Google Calendar so that could work pretty easily.

Its not that they couldnt do it. Its that they are expressely prohibited by apples patents. ( something about a unified search interface for your phone as well as the web )

Can you site a source here? That's an important distinction.

Here's an editorial that gives an overview of how the patent was relevant in the Apple v. Samsung case earlier this year, along with a link to the patent itself: http://www.androidpolice.com/2012/07/25/editorial-samsung-ha...

And never would allow that much integration across all apps. If that what you want, best get a Nexus 4.

I agree with everything you've said, but unfortunately, it still just won't be used on my iPhone. There's no way I will launch the Google app then click on the microphone icon. Replace Siri with this and I'll use it, but as long as it's a separate app on my phone it won't get used.

I will; I have an iPhone 4.

Sorry Siri... You've got a new big brother in my opinion.

Woah.. this may be one of the few times so many hyperbolic adjectives in a HN post seemed justified ;)

Just fired it up now, and can attest it simply blows Siri's UX right out of the water.

Kudo's to the Google team!

The voice recognition is outstanding but every query I'm trying (sports, news, weather) is returning text results.

In a situation where I'm using voice recognition I need it to read the response back to me, without that it's next to useless.

Am I missing something? Or is it the things I'm asking?

I think you have to phrase it in the form of a question in order to get a voice response.

Things like "How many meters in a mile", "Did the Tigers win?" and "What is the weather like?" all came up with an audible reply for me.

Try something with a one-shot answer like "When was the constitution of Argentina ratified"? It should speak assuming the latest Google app version.

I've tried weather in Glasgow, the time in San Francisco and a bunch of other very simple stuff.

I believe for the app to talk back, your language settings have to be set to "US English". That's how it is on Android.

Seems to be this.

Which is odd. I see no reason why English UK (or "actual" English as I like to think of it ;-) ) couldn't use English UK for recognition but speak back as if it were English US.

As it is it becomes unworkable set to English US it doesn't recognise stuff at all well with my accent, and English UK doesn't read stuff back which means that it's of massively limited use.

So great if you're in the US, for the rest of us, still a (very impressive) work in progress.

Well, it's great for search, but things that people use Siri mostly for are "please call X", "please remind me to do Y" or "make an appointment for Z".

As long as the Google app cannot interface with Contacts, Reminders or Calendar, this is just an interesting tech demo and nowhere near a Siri replacement.

Exactly. I am amazed that so few people get this. It's not about speed or accuracy, but rather about being tightly integrated with other apps on the phone.

+1. Voice recognition is mainly about hands-free utility; interacting with the phone's features and functions. Google's implementation is impressive, but it will just be something fun to show off for now.

> This is Next Level Shit. This is absolutely next level execution.

Crashed the first time I tried it.

And thus a million server blades evaporated into the night, taking my precious opinion with them. Reboot your fucking phone and try again.

Reply on news.ycombinator.com