Brits: Could you try something?

[Marvster]

I just heard about an amazing new service in the U.K. called Shazam. They say you can dial a number, hold your mobile phone up to a piece of music for 15 seconds, then receive a text message with the name of the track and artist. I've been skeptical of the audio fingerprinting technology that would make this level of recognition possible. I think it's free, so would any of our British forum friends be willing to give this a try and report back? The site is here: http://www.shazam.com Cheers, Marv

[Iaian]

Wow, that would solve the get pissed at the radio and myself for not remembering the name of that band problem. But it could be a serious problem here in America seeing that 75% of the population does not know how to drive without killing someone.

[whitefang]

Is this anything like playing vinyl records backwards to get our "message" from Satan? Would this work as well? And Iaian; a large part of the problem with not getting the song or artist name around here is the DJ's practice of not mentioning who or what was played in the middle of their "blocks"! This has gone on for years. Obviously assuming we all sit around all day, 24/7, listening to the radio and should thereby already KNOW! Whitefang

[spokenward]

more info on the service here: http://www.redherring.com/vc/2002/0509/2759.html and here: http://www.redherring.com/mag/issue101/1140019914.html The second mentions Sony's "eMarker". This appears to be abandoned. Another kind of neat idea. You hear a song - you punch a button on your keyfob - and when you get home it looks that song up on a server. Pat

[flyscots]

good idea, but what happens if the song is on the radio, you wait until it finishes to hear who it was and they dont say. Now you dont know and you cant do the 15 second thing. It will be difficult for this to be worthwhile. John

[Marvster]

[quote]Originally posted by flyscots: [b]good idea, but what happens if the song is on the radio, you wait until it finishes to hear who it was and they dont say. Now you dont know and you cant do the 15 second thing. It will be difficult for this to be worthwhile.[/b][/quote]Hey John, I'm less interested in whether the business model is viable than whether the technology actually works. Since you're in the right neighborhood, would you be able to give it a try? Marv

[flyscots]

yeah me too. actually I'm not in the right neighborhood, I'm actually in Philadelphia! I have e-mailed some people back home to see if they'll give it a try for us, which I'm sure they will, I'll let you know when they get back to me. John

[flyscots]

Marv [b]The results are in! [/b] :) My sister was kind enough to try Shazam a few times for me. Thanks Jen :thu: Sorry for the wait but here you go. The website says it costs 50pence (about 75cents) per call but Jen was only charged 33p each time. You call the number and are greeted with somethinmg like: [b] "Welcome to Shazam. We will tag your music... Hold the phone up to the speaker after the tone" [/b] Jen followed instructions and after 15 seconds Shazam hung up automatically. Just a few seconds later she received a text message saying: [b] "Shazam tagged Vex Red performing "Can't Smile" C ur tags and buy the CD at shazam.com" [/b] Interestingly, if you have a mobile (cell) phone close to a speaker you get some interference and the audio is affected. This DID happen when Jen called Shazam but it was no problem at all. I'm intrigued as to how Shazam recognizes the audio. I asked my sister to try to trick Shazam and the kind hearted sole of course obliged! Rather than playing music from a CD or the radio she played a downloaded mp3 (ssshhhh!) directly from the computer. Shazam succeeded. Then she tried to trick it by playing a cover version, again straight from a downloaded computer file. Again, Shazam pulled through. Clever little sister then attempted to play an untitled bonus track from the Deftones album Adrenaline. Shazam simply replied: [b] "Shazam tagged Deftones performing "Bonus Track" C ur tags and buy the CD at shazam.com" [/b] I'm intrigued, I'd like to know how this works, any ideas anyone? Now Marvster, me thinks Shazam owe us some commission payment! :D Jen told her friends, who told theirs and son on. Everyone is trying it, including the Dad of one of her friends who had all his workmates trying it! For the most part, everyone has been trying to trick the system, there is only ONE report of success. Shazam thought that it was the band 'Lay Down The Law' performing "500 miles" when it was infact 'Less Than Jake' covering the same song. So...very interesting, very clever, but not flawless! John Scotsman

[lakeside studios]

Damn! Another newfangled electronic service for me to get hooked on.Beat Shazam!!Yeah!!! Gotta do it.Now where's my mobile and reading glasses....... :evil:

[Kendrix]

Try an instrumental and see if it works. They may be recognizing lyrics - or else they are matching the audio stream to some huge database. I cant imagine how they could build such a huge database - how could they keep it current?

[flyscots]

It cant be the lyrics Kendrix, that was found out with the successful recognition of a cover version. It has to be something in the audio stream, but does anyone know what exactly it would be? :confused: I thought playing off mp3 would maybe confuse it but obviously not. Their database must be huge. As I posted before I really dont see how they can afford to do this when you think of the technology they must be using. Maybe they are just praying on people like my sister's friends and Mr Lakeside Studios - see Marv, where's our commission money!?!? ;) John Scotsman

[Marvster]

Hey John, Very cool! Please thank your sister profusely for going to the time and expense of testing the system. For those wondering how it works, while I don't have specifics I do know they're using audio fingerprinting technology that is, indeed, scrubbed against a huge database of clips. Several companies have been working on this for years -- it's the same thing that would've been used in a paid version of the late Napster -- so it's not difficult to believe they've compiled some hefty databases by this time. (Licensing must be pretty thorny, though.) The interesting thing is, you don't need the whole song in the DB. The technology takes snapshots/slices of the rhythmic and frequency content so that sometimes a fingerprint for an entire song takes up only a few kilobytes (I seem to remember hearing as low as 100). It all makes theoretical sense, but I'm still amazed it works so well. Wouldn't be too hard to think about applications for the same type of technology in music making. Thanks again! Marv

[flyscots]

Hey Marv Thanks for your explanations, it will be interesting to keep an eye on this and see where it leads. Not entirely sure about this bit though: [quote] The interesting thing is, you don't need the whole song in the DB. The technology takes snapshots/slices of the rhythmic and frequency content so that sometimes a fingerprint for an entire song takes up only a few kilobytes [/quote]Two things questionable here. While this idea makes perfect sense in terms of file sizes I'm sure there are plenty of songs that would need more than just a short section to be recognized. Believe it or not there are still a few bands out there who use various time signature changes, dynamics and include many different frequencies etc. Obviously the majority of songs, especially the most popular (ie. those played on the radio) dont have many rythmic changes in them, so the short samples would work. However, so many songs are so similar in rythmic content that the technology would have to very precise when differentiating between songs. This is especially the case with identical cover versions. Lots of questions in the magnificent world of technology! John Scotsman

[Roto]

If not for John's sister confirming it, I would've said no way this is going to work. Dammit! I want one here to stump!

[Marvster]

[quote]Two things questionable here. While this idea makes perfect sense in terms of file sizes I'm sure there are plenty of songs that would need more than just a short section to be recognized. Believe it or not there are still a few bands out there who use various time signature changes, dynamics and include many different frequencies etc.[/quote]Sorry, I wasn't clear. They take tiny slices from sections [i]throughout[/i] the song, and those are kind of compiled/merged/whatevered into a file that serves as the fingerprint. Cheers, Marv

[Bunny Knutson]

It seems to me that if you are supposed to hold the phone up to the song for 15 seconds, then the fingerprinting technology would only need to take a very short sample from the original song at intervals of around 10 seconds or so. That way, they would be able to recognize the song from ANY point with the song (regardless of time sig, tempo, key, or tone) as long as you played AT LEAST 15 seconds of the tune. This method would also cut down on info storage space, as each song would be sliced into a very small file. I'm probably repeating what has already been stated here, but I'm just trying to figure this thing out, too. Seems like a cool technology, that I might use, were it available and affordable.

[StoneKnife]

The article linked above explains how they do it: [quote] Shazam's "magic" happens through a pattern recognition-software algorithm developed by Avery Wang, the company's chief scientist. The algorithm picks out the salient characteristics of the song, or its "fingerprint," and then matches that fingerprint against a music database. The company's algorithm pares the musical sample down to the barest information needed to make a positive identification, which speeds the processing time to get a match and minimizes storage requirements. For Shazam's service to be effective, it needs a database that contains most of the music that users are likely to hear in public: the modern urban soundtrack of shops, bars, clubs, and pubs. Shazam says its service could accurately identify most songs if its initial database contained 40,000 or so titles--about the same number of songs that could be found at one of the U.K.'s largest music stores, like a Virgin Megastore. The company is scanning tracks into the database according to sales figures, entering the most popular songs first. Given the surprisingly small number of tunes that are ubiquitous at any one time, first scanning in only the most popular music means that--despite having only some 10,000 songs in its database--the service can be extensively tested well before its official launch. The scanning will continue in order to keep Shazam up to date and to expand its database. [/quote]:idea:

[Sylver]

Hey John, Thanks for the test drive on that service. Pretty cool technology. Though I'm sure I'm much too cheap to use it if it comes to the states. Jack

[Kendrix]

If its recognition of the audio fingerprint then this seems inconsistent with the statemnent that it recognized a cover of a tune. Something is not hanging perfectly together here.