formsy.help

speech signal process tools

formsy [-bn.m][-rn][-an][-vn][-ln][-nn][-sn][-hn][-fn][-gn][-ef] f0 fb syn

Where "f0" is a file containing F0, VUV, and amplitude data, and "fb" contains
the formant frequencies and bandwidths.  The program "formant" produces
files of the required format.  These may be used as templates for
hand-generated or modified parameter files.  The "syn" file contains the
synthesis output.  The options are:
 -bn.m
   Instead of using the bandwidths from the "fb" file, bandwidths are
	synthesized from formant frequencies by the formula
		bandwidth = n + ( .m * frequency) [null].
 -vn
   The probability of voicing in the .f0 file is ignored and the value
	of n is used throughout the synthesis instead (0 <= n <= 1) [null].
 -ln
   The bandwidths from the "fb" file are constrained to be >= n.  The
	-b and -l options are mutually exclusive [null].
 -nn
   Only the first n formants in the "fb" file are used [null].
 -sn
   The synthesis frequency is set to nHz [10kHz].
 -hn
   n constant "higher-pole" formants are synthesized  [0].
 -an
   n is the factor by which formant frequencies are to be scaled [1.0]
 -fn
   The fundamental frequency from the "f0" file is multiplied by n [1.0].
 -rn
   The frame rate of both the "f0" and "fb" files are multiplied by n [1.0].
 -gn
   n is a boolean (1 or 0) which determines whether amplitude correction
	is applied after synthesis.  This tends to improve the quality
	in the voiced regions and degrade quality in unvoiced regions [0].
 -ef
   f is the name of a signal file to use for excitation [null].  If f is
	specified, the .f0 file must NOT be specified and the v, s, f, r and
	g options are not available; the synthesis frequency will be that
	of the excitation file.

Formsy is a simple n-formant synthesizer designed to help assess the
validity of formant and F0 estimates obtained from the program
"formant."  "Formsy" is not a serious attempt at quality synthesis and
should be used accordingly.  In particular, the voiced and unvoiced
excitation functions can be substantially improved and the amplitude
balance between voiced and unvoiced segments is not quite correct.
These problems, and the very poor voiced/unvoiced estimates obtained
from "formant" are being worked on.