Looped sample tuning bug

• Ro 3 0

  Message de Roland le …
  Hello!
  Polyphone has some problems at finding tuning for short looped samples - it doesn't show anything except the tuning curve. I think Polyphone should first expand the sample by looping it until we get enough sample data to correctly calculate the tuning. Could this be fixed? Thanks
  PS Sorry for my bad English
  Bez tytułu.png
• 57 0

  Message de Michael le …
  First of all, your english is fine! B)
  
  I think tuning for short loops doesn't work due to lack of information which can be read from it's harmonics. You have to find the pitch manually - then switching on the sinus-sound can help you do it.
• 409 0

  Message de Davy le …
  Hi Roland, you are right: if the loop is too short then the Fourier transform and the autocorrelation have not enough data.
  In that case I automatically increase the range around the small loop = > will be in version 2.3
  
  If the sample itself is too small however, a manual tuning will be required as mentioned by Michael.
• Ro 3 0

  Message de Roland le …
  I think it would be a better idea to duplicate the looped segment as many times as needed to make the FFT
• BO 276 13

  Message de bottrop le …
  i would like very much to see you duplicate a looped segment. call me when you start with it.
  regards bottrop
• 409 0

  Message de Davy le …
  The case you are speaking is when we have a very short loop in a sample and the problem if we duplicate this portion several times is that both the autocorrelation and the Fourier transform will tend to give a resulting frequency equal to the loop length.
  
  For example if we select a loop comprising 224 points with a sample rate of 44100 (as shown in your example), the length will be 5.08 ms and we will have a perfect correlation (since we duplicate the portion) at the frequency equal to 196.875 Hz.
  If we add un single point, a loop comprising 225 points will have a length of 5.10 ms and the perfect correlation will be 196 Hz. The difference between 196.875 Hz and 196 Hz is 8 cents of a semitone.
  
  All this to say that the result will be too sensitive to the number of points included in the small loop.
  I may not be aware of all mathematical tools but based on the small loop we provide, extending both limits (start and end) provides a simple and better base to study the harmonic content.
• Ro 3 0

  Message de Roland le …
  You are right, but if we have a sample with a short loop, the soundfont player will still duplicate this loop as many times as needed to make the sound. Hearable frequency will be very dependent on the number of points in the loop, so the correlation measurements will be perfectly valid. Also, knowing the frequency of the looped sample will make it easier to adjust the loop length.
• 409 0

  Message de Davy le …
  You are right too! We may define better what would be the purpose of the root key computation.
  
  My opinion is that tuning a sample must be independent from the loop, the later being used in the process just to give a hint to know approximately where the pitch computation must be done, a hint to locate a stable area. To me, finding a very short loop that wouldn't have the same pitch than the whole sample would be an error.

Connectez-vous ou inscrivez-vous pour participer à la discussion.