Every recommendation algorithm I've seen does one or both of two things. The first being staying extremely close to things I have already expressed an interest in - never broadening my horizons.
Nate4D - I'd love to run the analysis on your church band to see if we tell the difference from your drummer and a click track. Send me a URL to a recording and I'll generate the plot. Paul at echonest.com.
I wonder how well this will actually work. Audio Fingerprinting is designed to be insensitive to most 'naturally occuring' music distortions such as encoding artifacts, noise and changes in equalization, but I don't know of any audio fingerprinting system that will work well when faced with people who are actively trying to evade detection. It won't be too difficult for a properly motivated MySpace user to find a set of filters that can be applied to any song that will allow the song to get a unique fingerprint, without actually changing how the song sounds. A quick trip through Audacity to apply a micro-pitch change, a little equalization, and perhaps a slight tempo change will probably do the trick. Of course, the folks over at Gracenote are pretty smart and may be able to adapt to evasions, but this will no doubt lead to even more sophisticated evasions. In the end I don't think it is possible to create a fingerprinting system that will be able to deal with people who are actively evading the system. In the end, the evaders will win.
Eric... thanks for chiming in... it is very interesting stuff, and I'm very curious about the process. How long does it take for a musicologist to rate a song on all 400 characteristics? How many songs can one person do in a day? What do you do about heterogenous songs (aka the 'bohemian rhapsody problem')? It seems to me that one of the great benefits of a music recommender is helping people get away from the U2s and the Coldplays and into the deep and interesting long tail, but to do that requires analysis of millions and millions songs. I wonder if the manual process that you guys use will scale enough to bring people to the long tail. Thoughts?
There's a nice table of OMR programs (some free, some commercial) maintained by Don Byrd of the School of Music at Indiana University: OMR Systems.
For fun, Don also maintains the Extremes of Conventional Music Notation where he records the extremes found in written music. Some interesting excerpted tidbits:
softestpppppppp (8 p's) in Ligeti's Etudes for Piano, 1st Book
loudestffffffff (8 f's) in Ligeti: Etudes for Piano, 2nd Book, (the 1812 overture only reaches ffff)
Instruments to be played by one performer in a piece - *Mahler: Symphony no. 5 calls for one clarinetist playing six different instruments.
Most repeated notes in a melody - 32 in Prokofieff: Toccata, Op. 11 (1912)
Imagine a world where if you didn't legally work for Apple, you couldn't write a program for their computer.
So.. what do you have to do if you want to write a program for your iPod?
Tolkien was dogged throughout his career by the literary critics that felt that the Hobbit and the Lord of the Rings was beneath the dignity of an Oxford professor (and certainly not worthy of their attention let alone praise). Perhaps Neal picks the term 'Beowulf' as a nod to Tolkien who was the preeminent Beowulf translator and scholar of the 20th century.
One thing I learned from this video: Never set up your demo next to the bagpipe-playing robot. player... Also... never attempt to build a bagpipe-playing robot.
For an open-source speech recognition system with a real open source licence check out the CMU Sphinx Project, a family of speech recognition engines, training tools and associated acoustic and language models. The latest version Sphinx-4 is written in Java and is released under a BSD-style license.
This paper describes the performance issues we encountered when developing FreeTTS. I think it is a pretty good representation of the issues involved in developing a high-performance Java application along with a comparision between a Java and a native-C version of the same application. This paper describes how we ported a native-C synthesizer (Flite) to Java (FreeTTS) and how were able to get better performance from our engine.
This is not a toy application but a real application that performs well in a domain where performance really matters.
The RNDTXT project is all about creating art from the random words that appear in SPAM. I especially like Vicimus GEGAN an ambiant piece created with Python, cSound, sox and FreeTTS.
I agree with you.. but in 1981 people were saying the same thing... the Apple ][ has 70% of the market share, there is no way that IBM with its 'PC' will be able to make up that deficit.
Re:The flagship...
on
D&D Is 30
·
· Score: 4, Funny
BTW, Unrounded D4s are definitely the most deadly. One of my most painful memories is stepping on one of those buggers.
The rules for scoring are such that the score for a submission is based upon the judges votes divided by the byte count. This favors smaller images. 'Simple' being only 72 bytes gets almost a 4 times advantage over 'agate'.
Here are the rules:
Each voter will choose their six favourite images based upon artistic merit. A first choice will get 6 points, the second will get 5 points, and so on.
The gold place winner will be determined by dividing the total number of points awarded by the byte count.
The silver place winner will be the entry with the highest number of points.
The bronze place will go to the entry with the highest number of points divided by the square of the bytes used, this rewards the lower byte counts while still requiring an interesting interesting image.
I think the EFF is mentioned because John Gilmore, one of the founders of the EFF (and Sun Microsystems employee number 5) is filing in amicus
brief . Mr. Gilmore has been fighting for the Freedom to Travel . A similar case.
So the real questions are: do you sing the songs when you read lotr aloud? Did you ever skip-ahead during a particulary long and seemingly non-essential song or section?
Slashdot Mirror
Every recommendation algorithm I've seen does one or both of two things. The first being staying extremely close to things I have already expressed an interest in - never broadening my horizons.
clearly you haven't used the wreckommender: http://wreckommender.com/
Nate4D - I'd love to run the analysis on your church band to see if we tell the difference from your drummer and a click track. Send me a URL to a recording and I'll generate the plot. Paul at echonest.com.
From a researcher at Last.fm: http://mir-research.blogspot.com/2008/09/ismir-2008-demos.html
A researcher from strands: http://www.scwn.net/2008/09/ismir-past-present-and-future/
And lots of posts from my blog:
I wonder how well this will actually work. Audio Fingerprinting is designed to be insensitive to most 'naturally occuring' music distortions such as encoding artifacts, noise and changes in equalization, but I don't know of any audio fingerprinting system that will work well when faced with people who are actively trying to evade detection. It won't be too difficult for a properly motivated MySpace user to find a set of filters that can be applied to any song that will allow the song to get a unique fingerprint, without actually changing how the song sounds. A quick trip through Audacity to apply a micro-pitch change, a little equalization, and perhaps a slight tempo change will probably do the trick. Of course, the folks over at Gracenote are pretty smart and may be able to adapt to evasions, but this will no doubt lead to even more sophisticated evasions. In the end I don't think it is possible to create a fingerprinting system that will be able to deal with people who are actively evading the system. In the end, the evaders will win.
Eric ... thanks for chiming in ... it is very interesting stuff, and I'm very curious about the process. How long does it take for a musicologist to rate a song on all 400 characteristics? How many songs can one person do in a day? What do you do about heterogenous songs (aka the 'bohemian rhapsody problem')? It seems to me that one of the great benefits of a music recommender is helping people get away from the U2s and the Coldplays and into the deep and interesting long tail, but to do that requires analysis of millions and millions songs. I wonder if the manual process that you guys use will scale enough to bring people to the long tail. Thoughts?
For fun, Don also maintains the Extremes of Conventional Music Notation where he records the extremes found in written music. Some interesting excerpted tidbits:
There are many others, quite interesting.
Wouldn't this technology be great for fixing up all those ID3 tags? MusicBrainz
... my father was a colonel.
I only have one song on my iPod too. It's by Led Zeppelin. When I hit shuffle play the song remains the same.
Imagine a world where if you didn't legally work for Apple, you couldn't write a program for their computer. .. what do you have to do if you want to write a program for your iPod?
So
Today's FoxTrot seems particularly apropos.
Er ...you should check to see who makes the Opteron. Hint ... it's not Intel.
Tolkien was dogged throughout his career by the literary critics that felt that the Hobbit and the Lord of the Rings was beneath the dignity of an Oxford professor (and certainly not worthy of their attention let alone praise). Perhaps Neal picks the term 'Beowulf' as a nod to Tolkien who was the preeminent Beowulf translator and scholar of the 20th century.
"In order to set her straight, I had to let her know that the reason she'd never heard of me was because I was famous."
One thing I learned from this video: Never set up your demo next to the bagpipe-playing robot. player ... Also ... never attempt to build a bagpipe-playing robot.
For an open-source speech recognition system with a real open source licence check out the CMU Sphinx Project, a family of speech recognition engines, training tools and associated acoustic and language models. The latest version Sphinx-4 is written in Java and is released under a BSD-style license.
Yep ... my user number is prime too, hey what's going on, they're all prime!
I don't live in the UK, so I can't enter, otherwise I'd submit this picture of the Douglas Adams Memorial
This paper describes the performance issues we encountered when developing FreeTTS. I think it is a pretty good representation of the issues involved in developing a high-performance Java application along with a comparision between a Java and a native-C version of the same application. This paper describes how we ported a native-C synthesizer (Flite) to Java (FreeTTS) and how were able to get better performance from our engine.
This is not a toy application but a real application that performs well in a domain where performance really matters.
The RNDTXT project is all about creating art from the random words that appear in SPAM. I especially like Vicimus GEGAN an ambiant piece created with Python, cSound, sox and FreeTTS.
I agree with you .. but in 1981 people were saying the same thing ... the Apple ][ has 70% of the market share, there is no way that IBM with its 'PC' will be able to make up that deficit.
BTW, Unrounded D4s are definitely the most deadly. One of my most painful memories is stepping on one of those buggers.
The rules for scoring are such that the score for a submission is based upon the judges votes divided by the byte count. This favors smaller images. 'Simple' being only 72 bytes gets almost a 4 times advantage over 'agate'.
Here are the rules:
Each voter will choose their six favourite images based upon artistic merit. A first choice will get 6 points, the second will get 5 points, and so on.
The gold place winner will be determined by dividing the total number of points awarded by the byte count.
The silver place winner will be the entry with the highest number of points.
The bronze place will go to the entry with the highest number of points divided by the square of the bytes used, this rewards the lower byte counts while still requiring an interesting interesting image.
I think the EFF is mentioned because John Gilmore, one of the founders of the EFF (and Sun Microsystems employee number 5) is filing in amicus brief . Mr. Gilmore has been fighting for the Freedom to Travel . A similar case.
So the real questions are: do you sing the songs when you read lotr aloud? Did you ever skip-ahead during a particulary long and seemingly non-essential song or section?