Revision 193:390d3c69ab40

View differences:

README
12 12 
spaced logarithmically in frequency, rather than linearly. The output
13 13 
bins are therefore linearly spaced in terms of musical pitch.
14 14 

  		




  
  15
  
    
This implementation is reasonably fast and is causal, operating


  		




  
  16
  
    
block-by-block on the input, though with quite high latency depending


  		




  
  17
  
    
on the frequency range specified. By default it produces output at a


  		




  
  18
  
    
higher time resolution than some other implementations, using multiple


  		




  
  19
  
    
kernel atoms per time block. The inverse transform is approximate


  		




  
  20
  
    
rather than exact (see the paper cited below for details).


  		




  
  21
  
    

  		




  
  22
  
    

  		




  
  23
  
    
About this library


  		




  
  24
  
    
------------------


  		




  
  25
  
    

  		




  15
  26
  
    
This library provides:


  		




  16
  27
  
    

  		




  17
  28
  
    
 * Forward transform: time-domain to complex Constant-Q bins


  		




  ......




  28
  39
  
    
 * Pitch chromagram obtained by folding a Constant-Q spectrogram


  		




  29
  40
  
    
   around into a single-octave range


  		




  30
  41
  
    

  		




  
  42
  
    

  		




  
  43
  
    
Building the library and plugin


  		




  
  44
  
    
-------------------------------


  		




  
  45
  
    

  		




  
  46
  
    
To compile this code, use "make -f <file>" where <file> is one of


  		




  
  47
  
    
Makefile.linux, Makefile.osx, Makefile.mingw32 depending on


  		




  
  48
  
    
platform. These files set up some flags and include Makefile.inc,


  		




  
  49
  
    
which defines the input files and so forth. You could equally write


  		




  
  50
  
    
your own Makefile which does the same.


  		




  
  51
  
    

  		




  
  52
  
    
The Vamp plugin part of the build expects to find a compiled version


  		




  
  53
  
    
of the Vamp Plugin SDK in a neighbouring folder (../vamp-plugin-sdk).


  		




  
  54
  
    
The unit tests also require Boost headers to be available.


  		




  
  55
  
    

  		




  
  56
  
    
Note that this code uses the KissFFT library (compiled from source


  		




  
  57
  
    
bundled here) in its non-default double-precision mode via the use of


  		




  
  58
  
    
the flag -Dkiss_fft_scalar=double in Makefile.inc. If you want to


  		




  
  59
  
    
build this code using an external KissFFT library or custom build


  		




  
  60
  
    
scripts, you may need to do some work to ensure you have the right


  		




  
  61
  
    
configuration.


  		




  
  62
  
    

  		




  
  63
  
    

  		




  
  64
  
    
Credits


  		




  
  65
  
    
-------


  		




  
  66
  
    

  		




  31
  67
  
    
The method is drawn from Christian Schörkhuber and Anssi Klapuri,


  		




  32
  68
  
    
"Constant-Q transform toolbox for music processing", SMC 2010. See the


  		




  33
  69
  
    
file CITATION for details. If you use this code in research work,


  		




  34
  70
  
    
please cite this paper.


  		




  35
  71
  
    

  		




  36
  
  
    
The implementation is reasonably fast and is causal, operating


  		




  37
  
  
    
block-by-block on the input, though with quite high latency


  		




  38
  
  
    
depending on the frequency range specified. By default it produces


  		




  39
  
  
    
output at a higher time resolution than some other implementations,


  		




  40
  
  
    
using multiple kernel atoms per time block. The inverse transform is


  		




  41
  
  
    
approximate rather than exact (see the paper for details).


  		




  42
  
  
    

  		




  43
  72
  
    
The C++ implementation is by Chris Cannam, Copyright 2014-2015 Queen


  		




  44
  73
  
    
Mary, University of London.


  		




  45
  74
  
    

  		












Also available in:  Unified diff