Formants method

From STX Wiki
Jump to navigation Jump to search

This method computes the formant frequency (candidate) tracks of a speech signal. The algorithm first detects the local peaks of the smoothed signal spectra (cepstrum or lpc) and then (optionally) applies a tracking algorithm to the peak array. The result is a set of parameter functions FOF(t,i) (with: i = formant candidate index, from 0 to n-1) which can be saved in the DataSet.

V2 method formants cepstrum smoothing.png

frq.
Defines the formant frequency search (fmin/fmax) and display range in Hz.
amp.
Defines the threshold amplitudes (amin/amax) for peak detection in dB. The line between the points (fmin,amin) and (fmax,amax) is used as the threshold function .
rel. to spg-maximum
The amplitude values are calculated relative to the spectrogram's maximum amplitude.
number of formants
The maximum number of formant tracks to be computed. Note, that this does not mean that this number will be found!
frequency unit
The unit for formant frequency display. This field is ignored if the formant tracks are plotted in the spectrogram graph.
multiple colors
If checked, each formant is drawn in a different color (as defined in the color scheme's function line colors). Note that the first color is defined by the Line number.
method
Selects the type of spectrum to be used for peak detection. The methods cepstrum smoothing and lpc transfer function are implemented.
number of coefs.
Number of cepstrum or lpc coefficients. If this value is set to 0, the number of coefficients is set to the sampling frequency in kHz + 4 (e.g. sampling frequency = 16kHz, no. coefs = 20, sampling frequency = 44kHz, no. coefs. = 48).
preemphasis
The factor for the preemphasis filter when using the lpc method (Any value between 0 and 1).
tracking
If checked, the tracking algorithm is enabled, otherwise only peak detection is performed.
bias frequency
A bandwidth-like frequency value used in the tracking algorithm to assign peaks to a formant track.
min. duration
Minimum length of a part of a formant track in frames. All parts with a shorter duration are removed by the tracking algorithm.
voiced only
This feature is not yet implemented, but if checked, only formant tracks in voiced parts of the signal would be assigned.