according to heresay, some grad students were responsible for burning out the servos on early disklaviers by programming them to hit all 88 keys at once (naturally, its probably the first thing anyone would try given the opportunity...).
its a fitting problem all right but in polyphonic transcription the number of independent variables is non-trivial, the cost of evaluating a guess is very high, and there are tons of local minima due to harmonic equivalence, which, due to the curious nature of instrument physics (missing fundamentals, even/odd harmonic patterns, detuning, etc), often beat the true optimum if your fitness evaluation is not sophisticated enough. a genetic algorithm will converge about as fast as 1000 monkeys banging on typewriters.
IIRC many of his rolls were actually "enhanced" by his publishing company... extra bass lines added, etc., without his approval and sometimes in rather inappropriate ways.
And, while its true that a MIDI recording is not 100% true to the source, its also true that an audio recording is not 100% true. a lot of the dynamics and spatial acoustic information are lost in a stereo rendering due to microphone placement, compression, etc. -- one could argue that playback on a disklavier is better in at least some respects.
I don't know about you, but if I had a disklavier in my living room, I'd pay good money for MIDI recordings of great pianists.
sounds like he really *did* take a sledgehammer to the thing!
btw; if you use a synth on a computer its pretty easy to retune MIDI into just intonation or whatever.
I used to think about the possibility of a disklavier that could retune itself automatically. Theoretically it should work... but it might not be good for the strings / frame, etc, and the piano tunerss society would probably lobby Santorum to sponser a bill to make it illegal.
The "high def" MIDI the slashblurb refers to is probably what yamaha calls "Yamaha XP MIDI".
I can't find a technical spec on this right now, but its mostly backwards-compatible with MIDI plus the addition of a few extra details about the piano performance, e.g. key stroke depth, using the MIDI controller extensions. (However I'd be really suprised if it it was actually possible to determine key stroke depth from signal analysis of an old recording).
If they did their homework it would should also have a higher clockrate. MIDI is notorious for its poor time resolution with a clock of only 1khz -- and studies have shown that virtuoso pianists can control timing down to the sub msec range, so this is essential.
The article also beats around the bush on the polyphonic transcription issue -- but since these are classical pieces, score following seems like an obvious if not relatively easy way to do it.
try digitizing 100 hours of lecture from tape (an entire semester's worth). have fun, see ya in a couple of days. the same thing from an ipod is just a drag-n-drop.
but if the ipod existed before P2P sharing and the internet, no-one would give a rats ass how easy it is to make a copy, thus my original comments about ease of distribution in light of these technologies, not specifically about the ease of getting the recording onto a computer (unfortunately I think this point was misinterpreted subsequently -- maybe I did not make it sufficiently clear, or maybe its just slashdot)
Linux Kernel 2.6 is worse than 2.4 (empirically for our application here) - BAD. (less stable, slower, scheduler is less able to balance things properly
2.6 has several schedulers available... which ones have you used and how do they stack up?
each layer of redundancy reduces the odds of total system failure by an order of magnitude.
suppose server X has a projected downtime of 1 day, and server Y has a projected downtime of 2 days because it runs a crappy bloated OS which requires reboots every week. now suppose we want to run an online business that makes $10K per day, then Y will cost $10K more per year than X. Now suppose you make a two-way cluster with failover. estimated downtimes are now 6 and 24 minutes. so, 2Y will cost approximately $75 more than 2X. on profits of 3.5 million a year, thats nothing.
i used to think uptime was important, but now i know from practical experience that in any environment where it actually matters (enterprise systems) every service needs to have failover anyways, so the fact that you have to reboot for a security update is really a non-issue.
Slashdot Mirror
If some statements are part of a loop, then we gratuitously indent them.
Really? I always thought that was the ghost of RMS reaching out of emacs to strangle me...
(which I suspect is why you say the clock is 3000/3=1 khz)
yeath that is pretty much what I was getting at... thanks for the detailed explaination though, very informative.
according to heresay, some grad students were responsible for burning out the servos on early disklaviers by programming them to hit all 88 keys at once (naturally, its probably the first thing anyone would try given the opportunity...).
its a fitting problem all right but in polyphonic transcription the number of independent variables is non-trivial, the cost of evaluating a guess is very high, and there are tons of local minima due to harmonic equivalence, which, due to the curious nature of instrument physics (missing fundamentals, even/odd harmonic patterns, detuning, etc), often beat the true optimum if your fitness evaluation is not sophisticated enough. a genetic algorithm will converge about as fast as 1000 monkeys banging on typewriters.
IIRC many of his rolls were actually "enhanced" by his publishing company... extra bass lines added, etc., without his approval and sometimes in rather inappropriate ways.
And, while its true that a MIDI recording is not 100% true to the source, its also true that an audio recording is not 100% true. a lot of the dynamics and spatial acoustic information are lost in a stereo rendering due to microphone placement, compression, etc. -- one could argue that playback on a disklavier is better in at least some respects.
I don't know about you, but if I had a disklavier in my living room, I'd pay good money for MIDI recordings of great pianists.
he would go through several strings a month...
sounds like he really *did* take a sledgehammer to the thing!
btw; if you use a synth on a computer its pretty easy to retune MIDI into just intonation or whatever.
I used to think about the possibility of a disklavier that could retune itself automatically. Theoretically it should work... but it might not be good for the strings / frame, etc, and the piano tunerss society would probably lobby Santorum to sponser a bill to make it illegal.
...unless its a Victor Borge performance.
Microtiming is control of event timing, but not necessarily *absolute* timing., e.g. the inter-onset time between two near-coincident notes.
Virtuoso pianists (typically having 15-20 years of formal training starting at age 2-4 years) have exceptional control of microtiming.
Non-virtuosos are very sloppy when it comes to timing. Non-musicians are even worse. Each category is like an order-of-magnitude difference.
There are plenty of virtuoso pianists who are totally dead in the "emotion" department.
The "high def" MIDI the slashblurb refers to is probably what yamaha calls "Yamaha XP MIDI".
I can't find a technical spec on this right now, but its mostly backwards-compatible with MIDI plus the addition of a few extra details about the piano performance, e.g. key stroke depth, using the MIDI controller extensions. (However I'd be really suprised if it it was actually possible to determine key stroke depth from signal analysis of an old recording).
If they did their homework it would should also have a higher clockrate. MIDI is notorious for its poor time resolution with a clock of only 1khz -- and studies have shown that virtuoso pianists can control timing down to the sub msec range, so this is essential.
The article also beats around the bush on the polyphonic transcription issue -- but since these are classical pieces, score following seems like an obvious if not relatively easy way to do it.
try digitizing 100 hours of lecture from tape (an entire semester's worth). have fun, see ya in a couple of days. the same thing from an ipod is just a drag-n-drop.
but if the ipod existed before P2P sharing and the internet, no-one would give a rats ass how easy it is to make a copy, thus my original comments about ease of distribution in light of these technologies, not specifically about the ease of getting the recording onto a computer (unfortunately I think this point was misinterpreted subsequently -- maybe I did not make it sufficiently clear, or maybe its just slashdot)
making 10,000 tape-to-tape copies is at least an order of magnitude more effort than dragging a single file into your P2P shared folder.
there is no difference; its just more likely to be a problem now due to the fact that improper distribution is an order of magnitude easier.
;)
laws are like software... 90% of the requirements are unknown at the time of writing (and lawyers are like devious perl programmers...
ya might want to enable uploads.
130 K/sec here and 30 min. to go...
Linux Kernel 2.6 is worse than 2.4 (empirically for our application here) - BAD. (less stable, slower, scheduler is less able to balance things properly
2.6 has several schedulers available... which ones have you used and how do they stack up?
each layer of redundancy reduces the odds of total system failure by an order of magnitude.
suppose server X has a projected downtime of 1 day, and server Y has a projected downtime of 2 days because it runs a crappy bloated OS which requires reboots every week. now suppose we want to run an online business that makes $10K per day, then Y will cost $10K more per year than X. Now suppose you make a two-way cluster with failover. estimated downtimes are now 6 and 24 minutes. so, 2Y will cost approximately $75 more than 2X. on profits of 3.5 million a year, thats nothing.
its not a bandaid, its a solution.
i used to think uptime was important, but now i know from practical experience that in any environment where it actually matters (enterprise systems) every service needs to have failover anyways, so the fact that you have to reboot for a security update is really a non-issue.
encryption/decryption or any other large-number arithmetic (e.g., adding up the value of your mutual funds).
timestamps
thats too bad its only on CC. however I did find them in mp3 format a while back on *cough*usenet (a.b.audio-books, I think).
the feynman lectures on physics (recorded from his lectures at caltech) are excellent; and a good way to learn something.
Links in social networks are also of variable quality; so does this mean that the "six degrees" meme is merely wishful thinking?
drop kde, its not worth it. try using a true lightweight window manager like fluxbox or xfce.
The stock market is also chaotic but that does not stop anyone from using computer models to analyze it. Difficult, perhaps, but not pointless.
chaos theory says that systems have complex responses with respect to their input. you are saying that the system has NO response to its input.
spoken like a true mainstream-media baby.