Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Google’s “Conversational Search” Goes Live On Chrome (searchengineland.com)
115 points by dazbradbury on May 22, 2013 | hide | past | favorite | 71 comments


The tech that powers this is astounding. But am I the only one that really doubts this voice-powered future everyone seems to be aiming for? From Siri to the Xbox One and Google Glass, there seems to be an overall assumption that voice is the interface everyone will be using- but didn't we assume the same when dictation software first came out years ago? That we'd all be dictating our documents rather than typing them?

The only place voice control feels natural to me is when I'm in a room by myself. So that rules out the office, public transportation, and my home, except at rare moments. I'm actually quite happy about that- the last thing I want is to be at work surrounded by people talking to their computers.


Voice search is a natural extension (and time will tell if its an ideal one), but I think the thing that is most interesting with all of this is the steps these companies are making in natural language processing.



I expect that someday voice interfaces will be driven by subvocalization detection, so other people won't hear you operating a voice interface.


I think that the next step after the subvocalization detection will be brain-computer interfaces. Actually they might skip directly to that because it might be easier in some way if they are actually in your brain. Or maybe subvocalization detection using a BCI.

http://www.youtube.com/watch?v=B1iyTsdZ3NY Erin Jo Richey - Brain-Computer Interfaces and the Coming Wave of Mental Telepathy

http://www.engadget.com/2009/05/14/darpa-working-on-silent-t...


As I understand it, "subvocalization" usually refers to the voice in your head that you experience when you read, so the parent post may have been actually making this claim, albeit inadvertently.


The voice in your head is accompanied by movement of the muscles that you use to speak. This can be detected: http://en.wikipedia.org/wiki/Subvocal_recognition


In lot of ways I agree with you. When I got my first android phone I was so excited to use voice search and I was planning on never typing on my phone again. And a few years later I rarely ever use voice search. Even in the privacy of my own home, I never use it. Voice search works well and its fast but I feel much more comfortable typing it, I guess.


I've only used voice search a few times:

1) Look, shiny, let's play with it.

2) Oh god typing out this long thing manually is going to be painful. Let's voice and then do some hand-editing as necessary.

That's it.


I used to feel the exact same way about using any sort of voice coms. If anyone else was in the room with me it felt incredibly awkward. The pinnacle was when I decided to play a video game with my then girlfriend in the room and use Ventrilo. She tried to respond to everything I said, and I went as silent as possible.

Fast forward sometime and using voice com software on my computer feels as natural and second nature as using a phone to me. Skype, Vent/Mumble, Web Demos, etc all feel normal.

I imagine that those that stick with using it will find it pushes their comfort zone less. I would also make the prediction that as voice becomes more readily available that the rising generation will just get used to using it and won't really give it a second thought.


Agreed. Lip reading via the camera would be much nicer to use (although probably an order of magnitude more difficult). :)


Mind reading would be even better. Until then, writing words with a keyword is in many cases still the best way to provide input, without feeling awkward or without bothering others.

And speech isn't weird in the same way that touch-screens felt weird for some of us at first. Because touching the screen of your gadget does not annoy other people.

Also, did anybody notice how awful speech recognition is for non-Americans? And they just got in the habit of internationalizing things properly, now we're going back to square one, except for the fact that learning to pronounce things with an American accent is much harder than writing words in English, so devices will annoy the heck out of us non-Americans, even if we know English.



I think this implementation is just a necessary first step. Imagine combining this with technology that can detect subvocalizations -- that is the future.


No one is asking this, but why is this chrome only (Disclaimer: I am a Firefox user)? is there a specific part that is tied to the native browser code (microphone and speakers can be accessed though web languages)? [Note] the tone of the question is curious and not accusatory


There was a session at I/O called "More Awesome Web" which seemed to imply that a web audio API (which enables live microphone input) and a web speech API (which enables voice-recognition driven apps) are both currently only supported in Chrome. Here is a slide from that presentation: http://www.moreawesomeweb.com/#32 (just use the right arrow to progress through the slides).

If that's what's holding up speech recognition for other browsers then hopefully these APIs can become more widely supported, but at the same time I don't think the demand for browser speech recognition is there yet. It makes sense on your phone, but bending over your desktop or laptop in order to talk to your computer somehow doesn't seem right yet.


Well, yes and no. It would be possible through Firefox, but the feature (through WebRTC) is very recent, and low-level -- you'd have to go through a prompt, then they'd probably have to connect with sockets, do a request, and come back with the final results.

Conversely, Chrome has had x-webkit-speech -- a high-level text-to-speech feature -- since 2011. The code's already all there, everything just needed to be linked up and presented a little better.

They'll probably come out with Firefox support in a while -- it just takes more work.


Me: "What is the airspeed velocity of an unladen swallow"

Google: "Did you mean airspeed velocity of an unladen European swallow"

Me: "Yes"

Google "According to Wikipedia, Yes are an English rock band who achieved success with their progressive, art and symphonic style of music."

D'oh. If you're going to drop in an easter egg that asks a question, assume the user who finds it will actually answer it.


Despite numerous attempts I couldn't get it to recognise my british accent saying 'Air speed' as anything other than 'Ass speed' or 'add speed'


Any time I said "four" it heard it as "full". I can't really pronounce Rs, so ended up with "fouurrrrrr" before it recognised it.


Oddly, it seems more accurate on my phone through Google Search/Now.


Are you logged into your Google account on Chrome? I know that there's voice personalization on the phone versions. I'd expect them to bring that to the browser, too.



Remarkable. Chrome asked my permission to use my laptop's microphone and was then off to the races. I followed the examples in the post but added, "What is his dog's name?" The first result was the Wikipedia entry for Bo.

http://en.wikipedia.org/wiki/Bo_(dog)


Asked it "Who Made You?"

And Google Replied: "To paraphrase Carl Sagan: to create a computer program from scratch, one must first create the universe." :)


1.) Rasberry Pi 2.) USB sound card 3.) Analog mic splitter: (plus other things of course) -

Run small mic's and speakers to most commonly used rooms in house. Star Trek computer search enabled.

Missing link: send results to (x) screen.


.... why not just use a mobile phone clipped to your shirt (since that's all a star trek communicator is).


Speakers to go along with the mics as well, to hear the spoken response.


CastleOS has demos that appear to be successfully demonstrating that the Kinect makes a fairly cost effective voice receiver that can cover an entire room.


Do you have a link to this?

Edit: Nevermind, I read that as CentOS...


Not working for me. It displays "No Internet connection."

Edit: see http://crbug.com/242861


I think the service is being hit pretty hard. I'm getting the same.


I don't think that Google can be slashdotted ;)


It's not google.com. Chrome has speech recognition service that goes to google's servers. And that service is being hit hard, not google.com.


You're probably right. It's Chrome's speech recognition service that is having this issue. It's also affecting other sites.


From the reports that might be geographical.


I'm in India now and keep getting "No Internet connection" as well.


This is coreference analysis with voice input backed by an ontology. Google have done a good job pulling together these technologies.


I'm not exactly an expert on search machine technology, so I'd be grateful if someone was so kind to help me out here:

I did a "conversational search" asking "what's an amazing voice" and Google returned a definition with a Youtube video of a woman singing. One of the comments simply says: "Amazing voice!" Why is this listed as a definition? What's happening here?


The comment defines that video as an amazing voice. Google's knowledge graph has included this opinion.


The conversational aspect was one of the differences between Google Voice Search and Siri (though it doesn't matter if you have conversation abilities if you fundamentally don't work, but that's another debate). Interesting to see the progress in this space.


To me, this would feel more natural in something other than the search bar of Google. When we make a new search, the thought process is that it is a separate and new search.

Although we all know that searches are by no means stateless, they were separate, unrelated questions in my mind previous to this. This makes searching more like a conversation (hence the name) and means that the UI to me does not fit this style. There is no record of what you searched before this.

Imagine if your chat client was displayed in the same context. Google is only including it in this way so that it gets the huge usage of its search engine, but to me a more chatty UI would make this more obvious.


> To me, this would feel more natural in something other than the search bar of Google.

like in google glass?


Cute. Not something I'd ever use, but cute.

Overall it works, but it seems it can't pick up the word "one" and "reviews". Or it's me who can't speak those words right.

Kept getting "Xbox ones you" and "Xbox won the fuse"


I can do voice searches, but the conversational aspect (asking "how old is he?") isn't working, and I can't seem to get hotwording to either. Is this still rolling out?


I get this problem too. Understands stuff just fine but semantic and conversational stuff is broke.


same, updated chrome to newest version, single searches work as they should but the conversational aspect is missing


you might need to update chrome


Neat, but what's the practicality for the average user? Obviously there are universal benefits for the disabled, but what comes next?


Why is Google so afraid of making a standalone application to run on desktops and not in a browser?

Then you wouldn't need your browser open to use "OK Google"


> Why is Google so afraid of making a standalone application to run on desktops and not in a browser?

There is a big difference between "doesn't see the value in the added cost" and "is afraid of".

> Then you wouldn't need your browser open to use "OK Google"

Chrome supports running in the background on startup (and prefers to do so), so conceptually there is no reason anything that can be done in Chrome that doesn't require on-screen UI couldn't be done without needing any application "open" (at least, without the user taking the step of opening an application.)

Seems to me that, both from a strategic perspective and a UX perspective, that's a more natural progression for conversational search than a separate standalone desktop app.


Given how much work they've put into web apps and stuff like NaCl, I don't think Google sees requiring a browser as any sort of downside.


My Chrome is Version 26.0.1410.65 and "Google Chrome is up to date," but I don't see it when opening www.google.com. I'm in Malaysia.


Chrome 27 is out on Windows/Mac and seems to be required. You'll need to upgrade to 27 manually if you're on Linux or wait for your distribution.


You have to use Chrome 27.


Gr. I'm on Linux/dev/28 and no dice...


It'd be nice if queries starting with "go to..." would automatically go to the first result, ala I'm Feeling Lucky.


Thats cool... but i don't want to talk to my computer while in my cube. Useful in my car though.


"What's my name?"

"Where am I going?" (If you use Calendar.)

"Who founded Google?" (I just thought this was impressive.)


Heads up, the conversational aspect only works if you have Search History enabled.


Works fine for me, search history disabled.


Would someone tell me exactly what makes this feature "controversial"? I am seeing this exact same headline everywhere.

Is there any merit to it, or is it the standard linkbaity headline we've come to expect.


I misread it the first time, too. It's conversational.


>"controversial"? I am seeing this exact same headline everywhere.

I think you may be misreading it. The title says "conversational" not "controversial".


Conversational, not controversial


Did you misread the title?


Damn.


Try asking "What's your name?" or "How are you?"


Try "Beam me up, Scotty" :)


"who are you" is good too


Hah!

Also, "what's my name"




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: