Path: utzoo!utgpu!watserv1!watmath!att!att!emory!swrinde!zaphod.mps.ohio-state.edu!ub!uhura.cc.rochester.edu!rochester!cornell!wayner From: wayner@kama.cs.cornell.edu (Peter Wayner) Newsgroups: comp.ai Subject: Re: The AI Breakthough -- What It Will Be Like !!! Message-ID: <47847@cornell.UUCP> Date: 1 Nov 90 13:37:07 GMT References: <35244@cup.portal.com> <25207@uflorida.cis.ufl.EDU> <16141@csli.Stanford.EDU> Sender: nobody@cornell.UUCP Distribution: na Organization: Cornell Univ. CS Dept. Ithaca NY Lines: 47 weyand@csli.Stanford.EDU (Chris Weyand) writes: >In erich@near.cs.caltech.edu (Erich Schneider) writes: >>>>>>> On 31 Oct 90 17:23:10 GMT, rlp@beach.cis.ufl.edu (Trouble) said: >>+> Maybe it's a trivial example, but I'll call a computer "intelligent" when >>+> I can dial randomly across the radio, drop in on a song, and have it >>+> identify the type of music, artist, and song, and do it as quickly as >>+> a human being. Let's see how much processor power it takes to pull >>+> that trick. >>This is a trivial example, but I get your point. One point I would like to >>make is that "processor power" would be the only thing to needed to solve >>this problem. Given a list of all of the types of music/groups/songs a >>human knows, along with their characteristics, it's just an algorithmic >>process (i.e. a Turing machine, by Church's thesis) to perform the ID. >I kinda doubt that it's purely processing power. Maybe recognizing individual >songs could be done by some kind of wave-form pattern matching but this >obviously wouldn't work for identifying the type of music. Hah, this has been done and years ago at that. I saw a talk by a the guy. He had been working for the American Society of Composers, Authors(?) and Publishers (ASCAP) and trying to find a way to automate their auditing of radio stations. It turns out that every artist gets a few pennies when their song is played on the radio. It would be too much of an auditing nightmare to check and make sure that every station was keeping up, so ASCAP hoped to have a computer do this. The algorithm, as I remember it, involved severely filtering the signal with high and a low bandpass filters which made a very narrow hole for one narrow frequency range. Suddenly the multi-frequency bopping turned into a binary signal where on signified the presence of 4100 cycles per second sound and off signified the absence. (Or some other good frequency.) Then identification just turned into a string search problem which could be done relatively quickly. As I understand it, they didn't digitize all of music of the world, just the top 100 or something like that. They also arranged it so the system could be polled remotely by telephone and updated whenever Madonna came out with her newest hot stuff. Peter Wayner Department of Computer Science Cornell Univ. Ithaca, NY 14850 EMail:wayner@cs.cornell.edu Office: 607-255-9202 or 255-1008 Home: 116 Oak Ave, Ithaca, NY 14850 Phone: 607-277-6678