The process is essentially: (i) High-pass the signal, (ii) pre-emphasise, (iii) find residual after LPC analysis, (iv) find all local maxima, (v) perform a DP search for peak sequences.
This algorithm was originally used in the "epochs" program of ESPS.
-I Identify program name and version number.
-i item Select input item number.
-l Use LX signal as source. Some processing is modified for LX signals.
-f Use FX item to define voiced regions. Voiced regions are taken to be regions with a FX value greater than 0. The fxanal(SFS1) or fxrapt(SFS1) programs are a suitable source.
-t Use TR item to define voiced regions Voiced regions are taken to be regions where the track is greater than 0.5. The vdegree(SFS1) program is a suitable source.
-v Treat whole signal as voiced. Default is to apply a simple voicing decision based on energy and zero-crossing rate.
-p polarity Specify the signal polarity as positive (1) or negative (-1). By default the polarity is automatically detected. Specifying the polarity when known increases the speed of operation.
SPEECH Any speech signal
LX Any Lx signal
TX Larynx excitation points.
ESPS David Talkin and Derek Lin
SFS Mark Huckvale