• Register
  • Help
Results 1 to 3 of 3

Topic: Tuning of orchestral samples: use of artificial intelligence?

  1. #1

    Tuning of orchestral samples: use of artificial intelligence?

    It\'s time to face this up.

    The problem of tuning of orchestral samples has been touched by Bruce and others on Tascam forum (correct me if I\'m wrong).

    In short, what has been said is that a big role in the lack of realism of a midi orchestra is played by an incorrect tuning, in the sense that, while instruments like piano have a fixed tuning, other intruments can and DO modify the tuning of each note according to its context.

    It\'s not a secret that when you play a major chord on the piano the major 3rd is flat and that becomes clear when you play together with a violinist or a singer, for example (they often have to un-correct their intonation!)

    Try a simple strings ensemble mockup of a very short orchestral passage and listen to the difference with a recording. One of the first thing that musicians notice and lets them say \"it\'s fake!\" is the sinthy/dead character due to the not natural tuning (let all other parameters be correct: articulations, ambience, ect.). Actually, if you have a very well trained ear, you should just notice this artifact by playing simple chords on the keyboard whith a good strings patch: notes do not blend together like real strings.

    This is a very old problem which was brought up together with the concept of \"key\" (tonality).

    Now the problem in a sampler is that you could program a natural tuning which sounds good in a given key (say C major) but which would sound wrong in another (say F sharp minor).

    Another solution, that I tried once and invite you also give a try, is correcting intonation in real time using the pitch wheel (program it to respond slowly!!!). It\'s a very good listening exercise (and a funny experience!)

    The programming approach would of course be the most interesting one. But it demands some kind of artificial intelligence.

    The reason why I started this topic is that, while cooking pasta this evening(!), I tought maybe a neural networks approach could be tried.

    I had the chance to study Computer Sciences at University in parallel to my music activities before devoting myself entirerly to pianist career and was very passionate about artificial neural networks (ANNs) in connection to sound and speech.

    For those of you who don\'t know much about Neural Networks, there are basically two kinds of them: supervised and self-organizing. The difference essentially is that the first kind needs examples to learn how to respond; the second kind builds up by its own a representation of the training space. I would try using the first kind.

    In this case training examples would be recordings of the real thing (orchestra, solo violin, ...) and we should provide midi mockups of the same recordings. It would be good starting with a solo instrument and a midi mockup with a highly out-of-tune sampled instrument.

    It\'s not interesting for everybody to give details of the way the learning procedure of - say - a back-propagation NN can be adapted to our problem or what to expect as a result and how to use it. I will give more ideas and details if requested (and - most important - I\'m just starting speculating about it!!!)

    I just stop now waiting for feedback.



  2. #2

    Re: Tuning of orchestral samples: use of artificial intelligence?

    I\'ve thought about this a bit before, but never went any farther than that. I play and listen to woodwind and brass instruments all the time in the various bands I play in... also the choir. One of our assistant directors is very picky about intonation, but has simple methods to fix it. Now, I don\'t remember exactly, so this is just a pulled-out-of-me-arse example, but I\'m sure the real information is out there in many places. Basically, the notes, depending on which part of a chord they are, have a set value to change the tuning. IE, 3rd of a major chord, play it 10 cents sharp, 3rd of a minor chord, play it 10 cents flat. The 5ths, maj and minor 7ths also get a similar treatment, but I think it was more than 10 cents. I assume this is based from a tempered tuning. So, as a result, making a gig file with appropriate tunings for each key, wouldn\'t even do the trick, because the intonation for E as the third of a chord, and the intonation for E as the root of the chord, would be different.

    So in the end, this is just a long-winded \"Yup, I agree\" type thing.

  3. #3

    Re: Tuning of orchestral samples: use of artificial intelligence?

    Right, Aenigma.

    As you and me said it\'s impossible to make correct tuning part of the instrument itself. What I\'m talking about is some kind of post processing (be it at a midi and/or audio level) where tuning information would be corrected on a context base.


Go Back to forum

Tags for this Thread


Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts