Conference napalm::commusic_v1

    You mean you want to play a tape and get the score output by your PC?
    
    Ouch. If such software/hardware is available on the consumer market,
    I've not heard of it. (Would that I had!!)
    
    I'm real curious to see if anything pops up...
    
    Edd

    Well, not necessarily "the score", but at least all the notes of the
    melody, embedded if necessary in coutermelody or harmony.  
    
    I'm talking about capturing one or two instruments, not an orchestra:
    mostly traditional acoustic music from the UK, including the part that
    wound up living in places like the Cumberland Plateau and the Great
    Smokies. As with most trad music, a given performance will be very
    different to any printed score for the tune, and quite often the
    performance is the only record of the tune at all, or the only record
    under that name. I learn by ear, but have found that I get on much more
    quickly if I can *look* at what's going on at the same time I'm
    listening.  And I figure that since my little $60 Korg tuner can make a
    good try at figuring out what's going on, surely there must be boards
    who can *really* do the job.
    
    						=maggie 

    no
    the way people here (and Laurie Anderson) make scores is to play a MIDI
    controller, such as a keyboard, into a sequencer program on a computer,
    and which then can print out the score on a printer.
    You can't hack it up, either.  It's an extremely subtle and difficult
    unsolved problem.
    Although there are somewhat unreliable MIDI pitch followers that can
    follow one note at a time moderately well; they also would plug into
    the computer via MIDI.
    Tom

    I believe what you are asking for is not quite available as a
    commercial product yet. But I'd guess it's coming soon. I think I
    read/heard of some work in a university (MIT?) to automatically analyze
    a recording of music and produce its score. It seems to me that this
    would be a very difficult task computationally, so that early attempts
    will require computers far more powerful that a PC. But probably the
    researchers will find more efficient algorithms, and possibly build
    specialized computers for such a task.
    
    Now that I think about it, it may have been a Nova program where I saw
    this...

    Automatic conversion to a score is very difficult.  Automatic
    conversion to MIDI does not have some of those difficulties, but it
    still has some of them.  It depends a great deal on what kind of
    instruments are used and how much ambiance there is on the tape.
    
    Reverb and other processing muddies up the signal.  Plucked string
    instruments have noise all over the spectrum.  But flute-like sounds
    (recorders, pennywhistles) are relatively clean and give better
    results.
    
    If you got as far as a MIDI file, you could then use a MIDI editor to
    scale the timing and filter out the glitches.
    
    It takes more than a simple board to do this.  You need lots of
    spectral analyses, which are not fast.  Depending on how fast the music
    is playing you may need 10 to 20 analyses per second of music.  And
    depending on the pitch range of the instruments, may need 512-point
    analysis to be able to accurately resolve a semitone.  So for every
    minute of music you need to do 1200 Fast Fourier Transforms of 512
    points each.
    
    Now a DSP processor may be able to speed this up, if it can do the
    transforms.  I am not familiar enough with their capabilities.
    
    It *can* be done, but there are lots of limitations, and it depends a
    lot on the quality of the recording and the kinds of instruments used.

    
    Since apparently (.-1) knows some tricks, perhaps he can explain
    how a waveform that contains many various fundamentals and harmonics
    (read: 'taped' music) can be separated into 'instruments'.
    
    The problem of separating even a piano chord into it's component
    notes from a composite (read: all notes 'struck' at once) is 
    still not solved as far as I know. But then, Comp. Mus. Journal
    is my only source of state of the art advances. 
    
    Perhaps you have a better chance with 2 widely varying timbral 
    sources....bass drum and piccolo?,........but what music is that?
    
    So 'very difficult' sounds optomistic. 
    
    ron
    

    As I said, instruments with simple harmonic structure are easier to
    deal with.  A flute trio should give better results than a solo piano,
    for example.  (nonharmonic partials don't help either, unless you know
    real well where they are)
    
    No references I know of.  This is all based on my own observations.

    
    Well, we jumped on Maggie's 'problem' without really asking what it is
    that she wants to do...
    
    Recorded media-to-ADC-to-score isnt here yet. Ok.
    
    But what is it you want to 'score' actually? Why score it? To avoid
    buying sheet music? To analyze harmonic schtuff? To investigate
    the TIMING nuance (before/behind the fraction of beat) of a certain
    performance?
    
    That in itself is a note...ie, what do you want 'technology' to do...
    
    ron
    
    
    

ron,
she mentioned in .2 what she wants to do

chad

    
    yep...that didnt pin it down enuf for me, since all performances
    will vary from sheet music.
    
    I guess the need is for a learning tool...a way to 'figure out' what
    a given phrase or solo was...in order to see what the musician
    played...but I'd argue that you'd spend less time training your
    ears to do same than writing the software. Solfege courses at
    a school will help. Imitating your fave soloist will help.
    Experimenting will help.
    
    Ear-to-scoring is still the quickest available method of transcription.
    
    No technological rescue yet........harhhrhrhrrr...
    
    ron
    
    
    

    I appreciate the responses so far, though I take a rather jaundiced
    view of your contention, Ron, that training my ear would be easier: 
    you speak in ignorance both of my ear and of my engineering skills :-)
    
    Seriously, DSP chips such as that from TI (TMS 38130 I think, it's been
    awhile) are in wide use for FFTs and graphics engines.  Has no one
    really put on on a card for this purpose?  Again, my little $60 Korg
    tuner can do the job...it just can't do it fast enough and it doesn't
    produce nonvolatile output.  It would be nice to get the whole score,
    but maybe I shouldn't even have mentioned that since what I'll gladly
    "settle for" is something as primitive as putting the "Mississippi
    Valley Waltz" in and getting "GBGGBBGBDBEDBGGABDEDBGA..." out.
    
    You see, I have no problem following the rhythm or figuring out how to
    pick a tune, I just can't work out what notes are being played except
    by trying to match them note for note...which takes forEVer.  Visually
    I can discriminate very subtle differences, aurally I'm pretty hopeless
    and always have been.
                      

    ROn's actually pretty much on course. You're little Korg tuner gets
    a relatively "clean" signal over what's likely to be a longer length
    of time than can be expected when doing what you ask. The tuner also
    gets only one note at a time. Play a chord and see if it reacts as
    fast and/or accurately.
    
    Ear-training, while neither your choice or particularly quick, still
    seems to be the fastest method of transcription. I can't imagine
    alternatives to be TOO far off in the future tho...
    
    Edd

    This month's Keyboard magazine advertises a unit called the MIDI-MIC
    which claims to do what you're looking for in a roundabout way.
    It's apparently a pitch-to-midi converter, which would allow you
    to then convert the MIDI datums to a score (with yet more software).
    
    I've not seen or heard the device, but my guess is it wouldn't be 
    overly accurate. SWAG.
    
    Edd

    Those Midi-mic things will work with a few restrictions.  Their output
    is monophonic - I think they use filtering and zero-crossings to get
    the pitch.  The ones I have seen do not output much velocity
    information (one has only 3 velocity values it will use).  Of course if
    all you want is pitches and timings, then that is enough.  I don't
    think they will do the job if there were multiple parts going on.
    
    So what kind of music are we talking about here?  What instrumentation?

    The Midi-mic also advertises what it calls "chord mode", assumably to
    figure chords.
    
    Edd

    Sounding better [npi].  So where would I get a copy of Keyboard?, it
    isn't where I usually find mags.  And since as I've said I'm a complete
    zero at all this, would I need more than that midi-mic device?  What
    kind of dollars am I looking at all told here?
    
    (to whoever just asked about instrumentation...I presume you mean
    musical instruments?  if so: banjo & tin whistle mostly (not together). 
    Sometimes flute because there are good tunes translatable to the
    whistle, ditto fiddle for the banjo & whistle both.  Sometimes guitar
    for the banjo, but not that often...the repertoires tend to be
    different)
    
    						=maggie