I was wondering if anyone knows about any studies investigating the accuracy with which musicians are capable of transcribing polyphonic music? This could include any type of musical style (classical music, pop, jazz, ...).

I'm interested in questions like:

  • Is there a limit with regards to the number of concurrent voices?
  • What is the error rate when an exact transcription is required?

Anything in this area would really be helpful. But I would be particularly interested in quantitative results.

Thank you very much,


  • I’m not convinced there’s such a thing as an “exact” transcription. Just as part-of-speech tagging, you can likely find what a large amount of things people agree on, but there is no such thing as a single right answer for every piece.
    – Édouard
    Apr 14, 2014 at 15:12
  • Yes, although this is an interesting question, it is unlikely that you will find an answer with "quantitive-results". After all, the error-rate and limit-of-concurrent-voices are both dependent upon the skill of the person transcribing - which is developed through practise... Apr 14, 2014 at 15:27
  • You're right, Édouard, there's probably a lot of scope for discussion on what a 'correct' transcription of a piece may be. But I'm sure you could at least define a measure/score of "how good a transcription is", which for example applies certain penalties on missed notes, octave errors, notes assigned to the wrong voice etc.
    – koffer
    Apr 14, 2014 at 15:29
  • Bob, you're also right, it very much depends on the training. But I guess there could be ways of classifying people by their experience. And it would be possible to report what was the maximum achieved accuracy/score. If anyone knows about any qualitative instead of quantitative results, I'd happy to know about these as well.
    – koffer
    Apr 14, 2014 at 15:34

2 Answers 2


It might be worth considering that performing a score is a translation process, inherently… and transcribing music is--presumably--the "back-transaltion" of this process. Worse, we aren't just going from one language to another, but also changing mediums, as well. In any case, as with other examples of translations, the back translation is often going to have glaring re-(mis-?)interpretations of what might have been in the original score--sometimes quite laughable ones.

For example, what do you suppose the chances are that a transcriber of The Firebird might elect to choose the key of C flat, had he never seen the original score?

Also, one might consider that a lot of music has no actual "score," per se, to begin with. (For instance, Coltrane's solo in Giant Steps.) When someone makes the executive decisions necessary to turn something like that into 'golf clubs,' it is likely that something is going to be lost in translation, at least insofar as Coltrane's original conception was concerned. Perhaps he never intended to commit such a solo to paper at all! Perhaps he would turn over in his grave to find out that there were budding saxophonists out there even attempting a note-for-note copy of something that was meant to be freshly minted at every performance.


You're unlikely to find poor-quality transcriptions published simply because transcriptions are done in drafts - much like a painter sketching through a painting.

The accuracy of one person's sketches are of another fully-completed painting - much like an art forger - are either convincing or they're not. The best obviously fool everyone. In order to determine a transcription's accuracy, it would have to be compared to the original score - and even then, the original manuscript (if it still survives.)

Further, there is a matter of notation. If I notate something with quarter notes @ (1/4=) 60bpm, it will sound the same as half notes @ (1/4=) 120bpm. There again, you would have to make a distinction between what is aurally correct is what is notationally correct.

There are composers who have done transcriptions of entire orchestral works. So, the limit of human hearing has not yet been surpassed. Musicians have a keen way of taking very complicated things and breaking them down into manageable bits - no matter how complicated. Your research would have to focus on the amount of material a human being could notate / remember on a single pass through the music.

Certainly, you would also have to separate your data into transcriptions done by people with Absolute Pitch and ones done by people without Absolute Pitch.

As you can see, it is very difficult to answer a question that still needs to be further defined.

  • You are talking about the actual practise of transcribing a piece of music, and I agree that first sketching the score is the way a composer/arranger would probably go about this task. Nevertheless, in the end there has to be a finalised score (if the work is to be performed again) which I would denote as the 'transcription result'.
    – koffer
    Apr 15, 2014 at 9:53
  • And yes, there are issues with notation and performance (also: note lengths frequently differ between notation and performance, just think of a staccato crotchet which can be shorter than a legato quaver in the same tempo). But these are all issues that I would imagine could be addressed by allowing multiple results to be 'correct' or -as mentioned above- by defining a measure that takes these issues into account.
    – koffer
    Apr 15, 2014 at 9:53
  • I do believe that questions like these are valid questions to ask: How well does the transcription capture the notes that have actually been played? How many concurrent notes can a trained musician recognise? When does musical experience and guessing come into play? I'm not so much interested in the memory aspect (I think this has already been investigated from a cognitive point of view).
    – koffer
    Apr 15, 2014 at 9:54
  • @Koffer - my larger point was that because musicians transcribe in drafts, they work out the errors until the score is accurate. To answer your first question: the character / integrity of the music must be preserved. The 2nd question: see my answer. 3rd question: guessing comes with inexperience and goes away with experience. Apr 15, 2014 at 13:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.