Description Issue of MelBands

[raraz15]

In the description of MelBands there are missing and misleading explanations.

Description

This algorithm computes energy in mel bands of a spectrum. It applies a frequency-domain filterbank (MFCC FB-40, [1]), which consists of equal area triangular filters spaced according to the mel scale. The filterbank is normalized in such a way that the sum of coefficients for every filter equals one. It is recommended that the input "spectrum" be calculated by the Spectrum algorithm.

The problems:

1. The default filterbank is the HTK formulation (MFCC FB-24 HTK, [1]) not Slaney (MFCC FB-40, [1]). But in the description the later is provided as an explanation. And the rest of the details about the filters regarding the area normalization also belongs to MFCC FB-40.
2. The FB-40, and FB-24 refers to how many filters are used in the original formulation of HTK and Slaney, which were fixed. However, in essentia one can use create as many filters as required. So the numbers 40 and 24 can create confusion.

My suggestion:

Description

This algorithm computes the energies in the mel bands by applying a frequency-domain filterbank to the spectrum. The filterbank consists of triangular filters with center frequencies distributed on the mel scale. By default, the triangular filters have equal height at their center (Following MFCC FB-24 HTK [1] but you can change the number of filters with the numberBands parameter). Setting warpingFormula='slaneyMel' creates equal area triangular filters where the filters are normalized in a way that the sum of coefficients for every filter equals one (Following MFCC FB-40, [1]),.

It is recommended that the input "spectrum" be calculated by the Spectrum algorithm.