readme

speech signal process tools

@(#)README	3.25 9/23/98 ERL - README file for ESPS demo directory

This directory contains a few demo shell scripts that will illustrate
some ESPS programs and help you to figure out whether the installation
went correctly.  The following scripts are included (please run them in the
order shown below):

    showEnsemble -- a demo script that displays a useful set of speech
                    data displays (a waveform, an f0 display, a 
                    spectrogram, and two label files) given a common
                    file basename for a set of speech data.

    createfiles  -- creates files speech.rc, and speech.lsf;
                    these are needed by other demos.

    specdemo     -- computing reflection coefficients and power spectra.

    pipedemo     -- like specdemo, except shows how to use pipes.

    mlplotdemo   -- creates multi-line plot files with mlplot(1-ESPS)
                    leaving connected-line plots in directory plot/ and 
                    vertical-line plots in directory plotv/.  Note that
                    specdemo and pipedemo must be run before mlplotdemo.

    filtdemo     -- demo of notch-filter design, general filter design, 
                    and filtering.

    vqdemo       -- demo of vq codebook design, encoding, and classification.

    lpcdemo      -- demo of single-pulse LPC analysis, vector quantization,
                    and synthesis, and spectrogram computation.

    cepdemo      -- demo of cepstal-based deconvolution by using fftcep.

    syntdemo     -- demo script of the analysis/synthesis package new
		    to Version 5.0.  This uses xwaves.  Watch the shell
		    window for instructions to hit continue at various
		    points.

    setup_ratios -- designs filters needed for ratios and logratios scripts 

    ratios       -- script to compute power ratios in bands (takes arguments)

    logratios    -- script to compute log power ratios in bands
                    (takes arguments)

    Xmakeplots   -- For X window based systems.   Puts up each type of
		    ESPS plot and a spectrogram.  Run all other demos
		    first to created the required files.

You should look at each of these scripts and then run them.  If they
do not run properly to completion, contact [e-mail] and 
we'll help you figure out what's wrong.  The script "clean" removes 
all of the ESPS files generated by the demo scripts.  

showEnsemble is a little different from the rest of the demos.  This
script sets up a display of speech data files, and you may find it
useful in your day to day work if you generally work with speech
waveforms, f0 displays, and spectrograms, and you would like to label
some events in the signals.  The script is commented so that you can
relatively easily customize it to suit your own favorite display of
speech data ensembles (feel free to copy it and experiment with it!).
You run the script with the basename of a set of files.  For a full
demo of showing ensembles, run it with

  % showEnsemble poet canyon question

to show three ensembles: "poet", "canyon", and "question".  The first
time you run the script, you will create the f0 files and
spectrograms.  You can choose whether to save the spectrogram files or
not.  Be sure to keep the terminal window available so that you can
see the prompt messages.  You will probably have to bring the terminal
window forward to see it after the data displays cover it up.

A full display ensemble $BASENAME consists of these files:
      $BASENAME.sd	   ($SPEECH -- the speech waveform)
      $BASENAME.f0	   ($F0 -- fundamental frequency)
      $BASENAME.phones	   (an xlabel label file for labelling phonemes)
      $BASENAME.words	   (an xlabel label file for labelling words)
      $BASENAME.sd.fspec   ($SPECT -- a (wideband) spectrogram)

After running createfiles, you can run all of the other demo scripts
directly by just typing their names.  The setup_ratios script should
also be run directly, but it isn't intended to illustrate anything by
itself -- rather, its purpose is to design three bandpass filters
needed by the scripts ratios and logratios.  The purpose of these two
scripts is to illustrate tool building with ESPS programs (look at the
scripts).  The ratios script takes three arguments: a standard ESPS -r
option giving the range of points in the input file, an input SD file,
and an output file.  In fixed length frames, ratios computes the power
in each of three bands and the ratios of these powers.  The output
file is an ESPS FEA file, one record per frame, with fields for the
powers and power ratios.  The logratios script does the same thing
except that log power is computed.  If you have access to waves+, you
should give the waves+ command "add_espsf name ratios command ratios".
This will add the name "ratios" to the waves+ menu.  Selecting this
menu item will cause the ratios script to be run on the marked
segment, and the resulting FEA file to be displayed in a
time-synchronized window with the fields labeled.

After the demos run, look at some of the output files.  Use
psps(1-ESPS) to look at the ESPS files.  For example, try

    %psps signal.spec | more
or 
    %psps -a signal.spec | more

Similarly, try

    %comment -a signal.spec
    %comment -a psignal.spec

Personally, I like to pipe the output of psps to a paging
program -- you might like the following alias:

    %alias see 'psps -l \!* | more'    

Plot the output spectra from specdemo and pipedemo.  
For example, 

    %plotspec psignal.spec

will show the spectrum of the two sine waves plus noise.  

If the parameter- and common-processing feedback bothers you, turn it
off with 

    %setenv ESPS_VERBOSE 0

(assuming csh).  

If you have D/A on your system, use play (1-ESPS) to compare 

    speech.sd     - original speech
    speech.synt   - produced by single-pulse analysis-synthesis
    speech.qsynt  - same as speech.synt except spectral parameters 
                    quantized with size-16 VQ codebook

Remember, you can send electronic mail to us at:

    INTERNET:  [e-mail]
	 WWW:  http://www.entropic.com/support/