Smoothes the selected PowerCepstrogram by averaging cepstra. The smoothed PowerCepstrogram is the result of two separate steps. In the first step, cepsta are averaged across time. In the second step, cepstra are averaged across quefrency.
Time averaging window (s)
determines how many frames will used in the first step, averaging across time. The user-supplied value will be divided by the Cepstrograms's time step value (its dx). If numberOfFramesToAverage, the result of the division, turns out to be one or less, no averaging across time is performed. If numberOfFramesToAverage is larger than one and is even, one will be added. Each new cepstral frame will be the average of numberOfFramesToAverage frames of the input Cepstrogram. For example, if numberOfFramesToAverage turns out to be 5, then the j-th new cepstral frame is the result of averaging the 5 frames with indices j–2 , j–1, j, j+1 and j+2 for all frames j=3..numberOfFrames–2, i.e. besides frame j, the 2 frames on either side are used in the averaging. The numberOfFramesToAverage has to be uneven to allow for this symmetric behaviour.
Quefrency averaging window (s)
determines how many quefrency bins will be used for the averaging across quefrency step. The number of bins used in this step is the result of the division of the user-supplied value by the quefrency step value (the Cepstrogram's dy). If the result turns out to be one or less, no averaging across quefrencies is performed. If the resulting value is even, one will be added. If, for example, the result happens to be 3 then the value in quefrency bin k will be the average value of the values in quefrency bins k–1, k and k+1.
The following commands should reproduce the smoothing described in the Hillenbrand & Houde (1996) article, where they use a 20 ms (10 frame) time smoothing and a 1 ms (10 bin) quefrency smoothing.
selectObject ("Sound xxx")
To PowerCepstrogram: 0.041, 0.002, 5000.0
Smooth: 0.02, 0.001
© djmw, January 17, 2014