Cloud Atlas is Chordally Dense
December 17, 2012 5:30 PM Subscribe
How difficult is it to transcribe an orchestral piece (say, a digital recording) into an accurate paper score?
Can an experienced musician or composer do this easily? Is it tedious? How do they avoid errors? I'm vaguely aware there are computer tools to help with this. What are the tradeoffs—is it highly automated or does it still require human a to control and check?
I guess it would vary somewhat on the musical complexity, but let's say it's a piece of relatively modern film music. I tried on a whim to pick out the parts on a piano, but found that because in actual music the notes actually blend together, it got very mindboggling very quickly. It was tricky; sometimes what appeared to be one note was actually a slightly different note, etc.
Can an experienced musician or composer do this easily? Is it tedious? How do they avoid errors? I'm vaguely aware there are computer tools to help with this. What are the tradeoffs—is it highly automated or does it still require human a to control and check?
I guess it would vary somewhat on the musical complexity, but let's say it's a piece of relatively modern film music. I tried on a whim to pick out the parts on a piano, but found that because in actual music the notes actually blend together, it got very mindboggling very quickly. It was tricky; sometimes what appeared to be one note was actually a slightly different note, etc.
Best answer: Shoot, I mixed up the names of two products. PitchScope is the automated one, which is so time-consuming (to clean up its many errors) that people don't use it much, preferring Transcribe!, which isn't automated.
Here's a YouTube tutorial of someone using Transcribe!
posted by Sidhedevil at 5:45 PM on December 17, 2012
Here's a YouTube tutorial of someone using Transcribe!
posted by Sidhedevil at 5:45 PM on December 17, 2012
Best answer: I did some work on automatic transcription back in college (about 6 years ago) and it is not an easy problem to solve. I haven't kept up with current research/technology so I can't say how much improvement has been made since then, but efforts were focused on just extracting the primary melody from a piece of polyphonic audio with varying degrees of success depending on the type and complexity of the piece of music.
An automated system can easily extract a monophonic signal accurately (such as an a cappella voice or simple instrumental melody), but a polyphonic, complex piece of music (such as orchestral music) requires the application of Auditory Scene Analysis to seperate the instrumental sources. Most systems used some psychoacoustic modelling to try and do in software the same things your ear is doing, so if you have difficulty separating the instrumental sources, a machine will have an even harder time doing so.
All this to simply say that an experienced musician/transcriber will do the job manually a lot better than any automated system currently can.
posted by TwoWordReview at 6:38 PM on December 17, 2012 [3 favorites]
An automated system can easily extract a monophonic signal accurately (such as an a cappella voice or simple instrumental melody), but a polyphonic, complex piece of music (such as orchestral music) requires the application of Auditory Scene Analysis to seperate the instrumental sources. Most systems used some psychoacoustic modelling to try and do in software the same things your ear is doing, so if you have difficulty separating the instrumental sources, a machine will have an even harder time doing so.
All this to simply say that an experienced musician/transcriber will do the job manually a lot better than any automated system currently can.
posted by TwoWordReview at 6:38 PM on December 17, 2012 [3 favorites]
Best answer: It's pretty difficult. Speaking from the manual end of things, a transcriber might start by making a piano reduction that gets all the melodies and harmonies, and then listening again and again and again and again to be able to sort out and write in the instrumental parts. Eventual success will depend on the relative complexity of the score.
posted by daisystomper at 7:46 PM on December 17, 2012 [1 favorite]
posted by daisystomper at 7:46 PM on December 17, 2012 [1 favorite]
Best answer: I had to do this when getting my degree in music, and its really hard. Basically to learn we would start with a single line melody, and we were given the starting note. So for example our lecturer would play a note on the piano, tell us its A and then play the melody. Over and over.
Once we had started to get good at that, then there would be a key change (implied by voice leading since its just a single line). Then once we got down pat, adding a second line. This takes a lot of repetition to get good at. And we would continue from there. I only did this for a year so the most I did was 4 line music with pretty distinct separation and it was still a very long slow tortuous process.
Only one guy in our class would get it first or second time, and he had perfect pitch and we were all deeply envious.
posted by Admira at 9:53 PM on December 17, 2012 [1 favorite]
Once we had started to get good at that, then there would be a key change (implied by voice leading since its just a single line). Then once we got down pat, adding a second line. This takes a lot of repetition to get good at. And we would continue from there. I only did this for a year so the most I did was 4 line music with pretty distinct separation and it was still a very long slow tortuous process.
Only one guy in our class would get it first or second time, and he had perfect pitch and we were all deeply envious.
posted by Admira at 9:53 PM on December 17, 2012 [1 favorite]
Best answer: This spring, I made, for my own professional use, a transcription of a rather complex harpsichord piece (this arrangement of Bach's famous Ciaccona), using nothing but a CD player, a set of headphones and paper and pencil.
The process is indeed basically the one Admira describes: you know a few parameters (in the case of the Ciaccona that would be the length of the piece, the meter, and the fact that the harpsichordist who made the transcription had transposed it from d to g), and so you just begin.
Now a good harpsichord produces a pretty defined set of sounds; the instrument is in fact better suited for musical dictations (as describe in the answer above) than a piano. Still, as soon as you get into real music and fast stuff, transcribing by ear requires a rather well-honed set of listening, memorizing and deduction skills. It also takes some time; imagine just having to copy out a complex score by hand, add the extra difficulty of analyzing and memorizing the sounds to that. We're likely not talking about days, but weeks of work.
There are two main reasons that a truly faithful transcription of an orchestral score of some complexity is nearly impossible:
1) The ears hear a huge bunch of sounds, overtones and fundamentals, timbres etc.. Now the brain needs to sit down, sort of, and disentangle the sounds into known patterns. that's great as long as a sound is easily recognized: the top line of the strings, the bass line, an oboe solo, a chord of three horns...But what do the middle voices do? Who the heck plays the middle voices, even if you maybe find out the harmony and rhythm of what they play?
If there's some complexity, you simply can't tell any more whether, say, what you hear is the clarinet playing a line in this octave, or the bassoon in that octave.
So: harmonic reduction - if you've got the set of skills described by Admira, no problem - only somewhat time consuming. Orchestral score: very difficult, and no guarantee of ultimate accuracy because of the impossibility to disentangle overtones and fundamentals of the middle voices.
2) Speed and/or rhythmic complexity. The human mind has a top speed limit for hearing separate notes (That is where some software may be of use). Beyond that, we tend to listen to starting and target points in the hope that everything in between behaves a little.
Now interesting music nearly never just 'behaves' and that's where the problems begin. Listen to the passage in my example in the middle of the arpeggio section, where the artist introduces faster note values and some non-harmonic passing notes. While I am confident that I nailed it in the end (although one also has to take into account that the two hands play on two different manuals, interlinking - which makes it easier to play and more interesting to listen to, but more difficult to hear and notate), some of these bars had to be transcribed by deduction rather than by actually being able to hear every single note - because that's simply not possible.
Of course, the location where the recording was made and the recording technique may also throw a bit of dust into the mix. You can verify this by listening to orchestral music with a score in hand. Can you hear the oboes in the opening tutti of the Serkin/Szell recording of Brahm's second piano concerto? I bet not very often. I also bet that Szell whipped them into playing their lungs out - doesn't matter. The composer was out after an accumulated sound, the recording is hissy, most of the microphones are focused on the brass and the piano, which is mixed out too brilliantly, and there you are as a transcriber, left with a bucketful of best guesses...
posted by Namlit at 1:10 AM on December 18, 2012 [2 favorites]
The process is indeed basically the one Admira describes: you know a few parameters (in the case of the Ciaccona that would be the length of the piece, the meter, and the fact that the harpsichordist who made the transcription had transposed it from d to g), and so you just begin.
Now a good harpsichord produces a pretty defined set of sounds; the instrument is in fact better suited for musical dictations (as describe in the answer above) than a piano. Still, as soon as you get into real music and fast stuff, transcribing by ear requires a rather well-honed set of listening, memorizing and deduction skills. It also takes some time; imagine just having to copy out a complex score by hand, add the extra difficulty of analyzing and memorizing the sounds to that. We're likely not talking about days, but weeks of work.
There are two main reasons that a truly faithful transcription of an orchestral score of some complexity is nearly impossible:
1) The ears hear a huge bunch of sounds, overtones and fundamentals, timbres etc.. Now the brain needs to sit down, sort of, and disentangle the sounds into known patterns. that's great as long as a sound is easily recognized: the top line of the strings, the bass line, an oboe solo, a chord of three horns...But what do the middle voices do? Who the heck plays the middle voices, even if you maybe find out the harmony and rhythm of what they play?
If there's some complexity, you simply can't tell any more whether, say, what you hear is the clarinet playing a line in this octave, or the bassoon in that octave.
So: harmonic reduction - if you've got the set of skills described by Admira, no problem - only somewhat time consuming. Orchestral score: very difficult, and no guarantee of ultimate accuracy because of the impossibility to disentangle overtones and fundamentals of the middle voices.
2) Speed and/or rhythmic complexity. The human mind has a top speed limit for hearing separate notes (That is where some software may be of use). Beyond that, we tend to listen to starting and target points in the hope that everything in between behaves a little.
Now interesting music nearly never just 'behaves' and that's where the problems begin. Listen to the passage in my example in the middle of the arpeggio section, where the artist introduces faster note values and some non-harmonic passing notes. While I am confident that I nailed it in the end (although one also has to take into account that the two hands play on two different manuals, interlinking - which makes it easier to play and more interesting to listen to, but more difficult to hear and notate), some of these bars had to be transcribed by deduction rather than by actually being able to hear every single note - because that's simply not possible.
Of course, the location where the recording was made and the recording technique may also throw a bit of dust into the mix. You can verify this by listening to orchestral music with a score in hand. Can you hear the oboes in the opening tutti of the Serkin/Szell recording of Brahm's second piano concerto? I bet not very often. I also bet that Szell whipped them into playing their lungs out - doesn't matter. The composer was out after an accumulated sound, the recording is hissy, most of the microphones are focused on the brass and the piano, which is mixed out too brilliantly, and there you are as a transcriber, left with a bucketful of best guesses...
posted by Namlit at 1:10 AM on December 18, 2012 [2 favorites]
This thread is closed to new comments.
posted by Sidhedevil at 5:37 PM on December 17, 2012