Alias: null
Common Name: SpectralContrastFilter
Class: org. imirsel. m2k. filters. SpectralContrastFilter

Overview: This module takes as input, a magnitude or power spectrum in the range from omega = [0, pi] and filters the spectrum through Octave-scale non-overlapping filters. The ouput of these filters is used to ompute spectral contrast feature.

Detailed Description: A spectrum on the range omegae [0, pi] represented as a 1 dimensional double array is taken as input, as well as the sample rate. Low and high frequency bounds are set as parameters as a ratio to the sample rate. Therefore, these values should be in the range of [0.0, 0.5], to represent DC and the half sample rate, respectively. An Octave-scale filterbank with a fixed number of filters is constructed and applied to the inputs, Spectral COntrast feature is then calculated from the output of these filters producing a 1-dimensional double array output, whose dimensionality is twice the number of filters used.

Data Handling: The input data is not modified.

Inputs:
Input Magnitude Spectrum
[D
Magnitude Spectrum (Double array).

Sample Rate
java. lang. Integer
Sample Rate (int)

Outputs:
Spectral Contrast feature
[D
Spectral contrast feature (Double array).

Properties:
Low frequency edge as ratio of sample rate
Sets the lowest frequency edge of the lowest filter in the filterbank. The scale is in terms of ratio to sample rate, i.e 0.0 being DC and 0.5 being the Nyquist frequency (half sample rate). Default is 0.

High frequency edge as ratio of sample rate
Sets the highest frequency edge of the highest filter in the filterbank. The scale is in terms of ratio to sample rate, i.e 0.0 being DC and 0.5 being the Nyquist frequency (half sample rate). Default is 0.5 .

Spectral Contrast Alpha value
Sets the range around the top and bottom of the sorted filterbank output used to estimate the Peaks and Valleys of the Signal.

Use base 10 logs
Contols whether natural logs or base 10 logs are used to calculate spectral contrast feature.