Borrar filtros
Borrar filtros

Why my mel filters do not overlap on each other?

1 visualización (últimos 30 días)
cat0530
cat0530 el 18 de Mzo. de 2016
Editada: cat0530 el 21 de Mzo. de 2016
The 26 filters should overlap on each other. Could someone help me with this?
  1 comentario
Star Strider
Star Strider el 18 de Mzo. de 2016
I’ve never used or designed MEL filters. I refer you to: Mel Frequency Cepstral Coefficient (MFCC) tutorial since it seems to be a comprehensive discussion. Compare your code with that discussion.

Iniciar sesión para comentar.

Respuesta aceptada

Rick Rosson
Rick Rosson el 20 de Mzo. de 2016
Editada: Rick Rosson el 21 de Mzo. de 2016
I think the problem is in the second elseif statement:
elseif (k>=f_range(m+1) && k<=f_range(m))
These two sub-conditions are mutually exclusive, so this condition is always false. As a result, the line after this condition, which creates the right half of the triangle, is never executed.
Please try:
elseif (k>f_range(m) && k<=f_range(m+1))
That being said, there are much easier and more efficient ways to create the filter bank in MATLAB, for example:
  • logical indexing
  • the tripuls function
  2 comentarios
cat0530
cat0530 el 21 de Mzo. de 2016
Yes, you're right. I've changed the second statement and it works! Now the problem is that only 24 filters are shown in the graph instead of 26. I've tried changing the outer for loop but it says index out of bounds
Rick Rosson
Rick Rosson el 21 de Mzo. de 2016
You need to create 2 extra points in f_range. So replace the following two lines:
%Need 26 points spaced linearly between minfreq and maxfreq
f_range = linspace(minMel, maxMel, nfilts);
with these two:
%Need 28 points spaced linearly between minfreq and maxfreq
f_range = linspace(minMel, maxMel, nfilts+2);

Iniciar sesión para comentar.

Más respuestas (1)

Rick Rosson
Rick Rosson el 20 de Mzo. de 2016
Editada: Rick Rosson el 20 de Mzo. de 2016
Please review the tripuls function.
  2 comentarios
Star Strider
Star Strider el 20 de Mzo. de 2016
I’m not certain that will work. When I looked up the MEL filters (that are designed to emulate cochlear frequency responses), the pass-bands were intersecting and looked something like: ...XXXXX...
Rick Rosson
Rick Rosson el 20 de Mzo. de 2016
It will work. I've used it successfully.

Iniciar sesión para comentar.

Categorías

Más información sobre Spectral Measurements en Help Center y File Exchange.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by