ADSET1

ADSET1 - compute global spectral descriptor data set #1

Compute global spectral descriptor data set #1.

To each of the input spectra A0 to AN-1 the following processing is applied:

• The hearing threshold is applied to the input spectrum A (if input HTH is set to 1) and the frequency range is truncated to the selected range (FMIN, FMAX). The following spectral descriptors are computed from the linear spectrum:

• Energy (rms)

• Flux: the 1st order flux is computed, which is the correlation coefficient of two spectra. To compute a higher order flux the correlations must be multiplied.

• The amplitude spectrum is converted to dB (AREF -> 0dB) and the following spectral descriptors are computed from the logarithmic spectrum:

• Deviation: the standard deviation of the logarithmic amplitudes

• Smoothness: the sum of the differences of each log. spectral amplitude ai to the local average around ai.

• The value AFLOOR is subtracted from the logarithmic spectrum and all values of the resulting spectrum which are lower than zero are truncated (set to zero). The following spectral descriptors are computed from the shifted and truncated spectrum:

• Centroid and spread

• Linear spectral slope and octave-scale spectral slope: for both values the slope of the linear regression is used. For the linear slope the x-values are the frequencies in Hertz and for the octave slope the x-values are (relative) frequencies in octaves.

This SPAtom was developed for the NOIDESc project in 2006.

Usage:

ADSET1 N A AREF HTH DF FMIN FMAX AFLOOR

Inputs:

N

The number output vectors (frames).

A

The FFT amplitude spectrum (linear).

AREF

The reference amplitude (linear).

HTH

Hearing threshold. The following values are supported:

0 - dont' apply

1 - apply

DF

The FFT resolution in Hz.

FMIN,FMAX

The analysis frequency range in Hz.

AFLOOR

The amplitude floor for Fct/Fsp in dB. Note that AFLOOR is only used for the computation of Fct and Fsp.

Outputs:

RES1

RES1[*,0] - rms energy in dB (Arms)

RES1[*,1] - spectral centroid frequency in Hz (Fct)

RES1[*,2] - normalized spectral spread (no unit) (Fsp)

RES1[*,3] - spectral slope in dB / oct. (Osl)

RES1[*,4] - spectral flux (no unit) (Sfx)

RES1[*,5] - spectral smoothness in dB (Ssm)

RES1[*,6] - amplitude variance in dB (Avar)

RES1[*,7] - spectral slope in dB / kHz (Fsl)

All output values are computed for the specified frequency range (or the corresponding octave range).

See also

<SP-atoms>