Re: About the possibility of including automatic transcription into lilypond

[Graham Percival]

Gabriel Simões G. da Silva wrote:
> My name is Gabriel Simões and right now I'm starting my master degree in 
> electrical engineering, specializing in dsp and audio dsp.

I'm doing a Masters in computers science and music, so it's a highly 
related field.  I don't know what you mean by "partitures", though.

Automatic transcription is an interesting area; the general workflow is:
- partition audio into individual voices
- segment monophonic sound files into duration data.  (generally you 
quantize and round off values here; a sound that is 1000.3 ms long 
probably intended to be 1000 ms if you have 60 BPM)
- detect the pitch during each duration  (again, you can round off 
values.  If somebody is playing Western classical music and plays a 
frequency of 441 Hz, you can be certain that they intended to play 440 Hz)
- output list of pitches and durations.
- (optional) translate pitches and durations into Western score notation


If you begin with a monophonic signal, then it's not too hard -- 
although as you note, there are many details to take care of.  I'm 
currently working on the segmentation phase for query-by-humming instead 
of transcription, but it's the same problem.


> I'd like to know if the lilypond community have interest in this project, and 
> if yes, what should I consider ? including this into lilypond source or 
> developing a 3rd party tool?

3rd party tool.  Translating pitch/duration data into lilypond is a 
trivial task compared to the other parts.

Cheers,
- Graham