52 Comments
- nreynolds, on 10/12/2007, -1/+18no, sorry, it says *speech*-to-text. Made up rap lyrics won't translate.
- JimmyDushku, on 10/12/2007, -0/+16So I'll now understand what 50 Cent raps about??
- nreynolds, on 10/12/2007, -2/+11edit: digg screwed me in the comment system.
digg me left. - CapeKid, on 10/12/2007, -0/+7yeah just search for "reminds me of the time when"
- JamesWilson, on 10/12/2007, -1/+7http://www.podzinger.com/
- zachws, on 10/12/2007, -0/+4woah there kettle, you are on a mission from god. how furious i am at someone presenting something i cannot rationally dissect!
- eric1, on 10/12/2007, -0/+4I'm a little skeptical myself. If I search for Wii, why should I get any results? (Eg, why would Wii translate to Wii, and not Wee, etc... ) Does the software take into account the context in which the word is spoken? Is that even possible? Are the YouTube videos 'pre-processed'? If so, how often are they reprocessed? Youtubes archive is huge.
Great idea, but I'm not totally convinced yet. If it works like it seems it should, expect them to be bought by Google any day. - brad77, on 10/12/2007, -0/+3Did Podzinger know that the video was recorded at night? Is nighttime voice recognition more difficult?
- inactive, on 10/12/2007, -0/+2What do you mean you're little skeptical? This wordspotting technology is known and available for years already. How do you think the NSA sort your phone calls out?
In fact you can even do this stuff on your own computer if you are willing to shell out the big bucks, with ScanSoft MediaIndexer ( http://www.nuance.com/audiomining/ ) you can searching for spoken words in videos, their pricing is based on runtime licensing like the Oracle database. You know, ScanSoft, the same company who developed OmniPage which let you capture text in your scanned documents. They are not posting a hoax on their web site for the heck of it, this is the real deal. - jcact, on 10/12/2007, -0/+2second result for a search of "Nintendo"... "girls making out"
Amazing, somehow this search engine reads minds and knew what I was really searching for! - 5hinmyoken, on 10/12/2007, -0/+2i tried the podzinger _youtube_ search tho ;)
- kalleanka, on 10/12/2007, -0/+2So if a small start up have been able to build this technology now, is it possible that our government with way more resources had this technology 10 years ago? And is it possible they were and still are listening to all our phones?
Man, this is creepy. - inactive, on 10/12/2007, -0/+2Now for the list of most-often said words by youtube users in comments:
1.) fake
2.) *****
3.) *****
4.) *****
5.) americans
6.) brits
7.) are
8.) stupid
9.) fag
10.) noob - iliketurtles2, on 11/20/2008, -0/+2I just did a quick search for MacBook and it came up with this;
0:01:12 ... Okay on other subjects besides MacBook -- this is the feeling gay and I am going to be preparing my invoices. However I did not create
What he actually said...
Okay, on other subjects besides my book, this is the billing day and I am going to be preparing my invoices... - RedStateRetard, on 10/12/2007, -0/+1Is "naruto" #1?
- marshallk, on 10/12/2007, -0/+1The particular search result highlighted in the article is a good example of compensating for noise - Podzinger picked up the word "Starbucks" over the noise of a motorcycle ride through a city at night.
- JamesWilson, on 10/12/2007, -0/+1http://www.podzinger.com/results.jsp?scol=youtube&q=deny+holy+spirit&col=en-all-youtube-ep
- safiire, on 10/12/2007, -0/+1I think this is cool, but I haven't yet figured out if it is just finding things via youtube tags or if it's actually searching what people are saying.
So far it seems it could just be doing it by tags, with what I've found in searches so far. - gmprunner, on 10/12/2007, -0/+1How is this possible? Does the word have to be spoken clearly with no accent or background noise or anything? Arg, it's so weird...
- marshallk, on 10/12/2007, -0/+1Check out Podzinger does with the rest of the site, it doesn't just use meta data like tags. You can find examples too of searches where the word is not in tags just in the words used - try searching for "autumn" or "castration" for example. Pretty cool. And, if you search first for castration and then for "political castration" in quotes you'll see that it supports boolean search. Way cool.
Re time required, I imagine it's not 100% or anywhere near it right now - but if you look at what competitor Blinkx.com does with a large amount of video - they do speech to text to search on so much so fast! No YouTube there though. - Osmanthus, on 10/12/2007, -0/+1Microsoft shipped example code that did this in its early speech development kits many years ago; its not perfect, it doesnt get every single word matched, but its good enough for searching something like youtube. So the tech is nothing special.
- 47knight, on 10/12/2007, -0/+1how the hell is this possible?
- Junkey, on 10/12/2007, -0/+1Spooky.
- eric1, on 10/12/2007, -0/+1"Political castration" did work well, but try something like baseball. I didn't listen to the results, but just take a look at what you get. You can't tell me that baseball appears in all of those videos that come up. Just look at some of them.
To be fair though, some words did turn up some good results. - eric1, on 10/12/2007, -0/+1I agree; it's not obvious that it's actually searching the speech. I searched for 'Jamba' as in 'Jamba Juice', and it brought up a video description (that does not appear to be speech from the video) that includes (and highlights) the word 'Jamba'. It might say it in the video too, but so what?
I noticed when I searched some of the non Youtube content that it would give me a link to the exact instance of the word I searched for, but it seems that you'd find the word many more times in the same audio than the once it showed me.
The title is/seems misleading too -- the sheer amount of time to 'download' and process all of Youtube's content, from speech to text, is not at all trivial, and I doubt that Podzinger can offer that. - b05q, on 10/12/2007, -0/+1ufia:
the audio the phone companies/NSA work with is recorded by them, and aside from the potential crappiness of the microphones in cell phones is going to be great quality. it's also almost always one, isolated voice. youtube/TV/music/etc. are often lots of different people, all sorts of non-verbal noises and the audio quality is going to be necessarily horrible (to make it small enough to be web-viable). the more compressed a sound wave is the harder it is to analyze algorithmically.
no matter the quality of the recording, no one has figured out a way to isolate a single track out of a mixed recording (e.g. taking just the guitar solo out). you would have to be able to do something along those lines to effectively transcribe youtube.
better than i could do, it's just a VERY ambitious goal. - sockpuppets, on 10/12/2007, -0/+1Doesn't work, I searched ***** and got this. Not that I'm, um, complaining. :)
http://www.youtube.com/watch?v=9Lhv5qT4qdc - zachws, on 10/12/2007, -0/+1"How’s the quality? I did a search for Starbucks and Podzinger found 120 results so far - compared to 3,000+ on YouTube searching text. Those results are interesting though, including one video uploaded today of a man driving a motorcycle through Taiwan at night, past a place that reminds him of Starbucks. He says the name of the coffee chain at the 3:46 mark in the video, Podzinger shows us. That is very impressive!"
- jcact, on 10/12/2007, -0/+1with the superior youtube search engine FYI.
Not the one that seems to interpret OS X as OS Sex. - barthosch, on 10/12/2007, -0/+1Your search - here's the deal - did not match any podcasts.
- inactive, on 10/12/2007, -0/+1I searched for "Einstein" and it found a sentence "You know having an Einstein", but the actual words were "You're not having a nice time?" :)
- 5hinmyoken, on 10/12/2007, -0/+1Your search - ***** - did not match any podcasts.
- 5hop4orce, on 10/12/2007, -0/+1Firefox search engine plugin?
- kloof, on 10/12/2007, -0/+1No Diggnation results for "I'd take it in the arse" (or ass) :(
- drewolanoff, on 06/10/2008, -0/+1results seems really shaky.
- bking, on 10/12/2007, -0/+1If you know a line from a video and search for it (an important application for this, I'm sure), it doesn't work for *****. I tried for about 15 minutes, and I didn't find a damn thing. Boo.
- devlo, on 10/12/2007, -0/+0I'm with the skeptics, but not entirely. I tried out they audio recognition on audio podcast and results were'nt that satisfying (though clever app nonetheless). Now the YouTube search well... it seems as part of it scraped from the audio but i'd say about most of it is from the tags/description if not all of it. YouTube DO have an API for that. Still, as I said, great app.
- korupture, on 10/12/2007, -0/+0Doesn't work quite as well as described.
- nickboyett, on 10/12/2007, -0/+0This didn't work as well as i expected from reading the article. But if you are looking for videos on youtube or other popular shared video sites www.vdoogle.com is a very useful tool
- wankerface, on 10/12/2007, -0/+0"Results 1 - 10 of about 293 for *****"
- tamarind, on 10/12/2007, -0/+0Believe it!
- bmeckel, on 10/12/2007, -1/+1damn, now i can't search for the lyrics to the fast food freestyle! i love that thing (goes and plays it and sings along) ok i'm done with my shamefullness for the day
- Kranklin, on 10/12/2007, -1/+1Amazing! Search for "Historic day in washington" and see what you find
- wankerface, on 10/12/2007, -1/+1Still a few kinks to work out:
http://xs.to/xs.php?h=xs511&d=07014&f=podzinger_pnis.jpg
Yes, I found that doing a search for "penis"...
The speech-to-text in general seems pretty abysmal, although I could definitely see this being very useful for certain searches (just not for transcripts). - kettle, on 10/12/2007, -2/+2I call *****. The state-of-the art in speech-to-text is not yet up to this, at least not in automated form, and certainly not for the great majority of noisy environment, open-domain, unlimited vocabulary scenarios. This is hype. I do research in this field and have yet to see anything even remotely approaching the claims that this appears to be making (if they aren't really claiming to have accurately, automatically transcribed all the videos on youtube, then I apologize).
- DelMonte, on 10/12/2007, -1/+1Maybe it's too powerful and that motorcycle was actually driving by a person that said "Starbucks"?
- Zapcome, on 10/12/2007, -0/+0What a good way to find all the Family Guy Episodes on Youtube!
argh!...I guess this would be the end of all the copyrighted content ... :'( - echoforever, on 10/12/2007, -0/+0"the ability to search audio in YouTube is going to help raise the profile of Podzinger in a big way"
This will eventually provide them with enough resources to work on rejecting background noise, include search based on other languages etc. In no time, the search using Podzinger will yield results that match the results based on YouTube's search facility. - tonich03, on 10/12/2007, -0/+0they porbably have monkeys transcripting all the videoz
- bmeckel, on 10/12/2007, -2/+1sorry, i hit the wrong reply *button*
(whispers) crispy say crispy (/) CRISPY -
Show 51 - 52 of 52 discussions



What is Digg?
Digg is coming to a city (and computer) near you! Check out all the details on our