This code is distributed mainly to demonstrate how SNR is computed
for speech (or audio) signals. The installation procedure described
here is not 100% reliable, but should work across most environments
with some simple changes.

Let's assume we want to install the code in a
directory named /u0/smith/, and you have downloaded
the tar file into $HOME/snr.tar.gz.

To build this code, you will need access to GNU make.
On some systems, this is installed as "make". On other systems,
it is installed as "gmake". "make --version" is an easy
way to tell which version it is. In the instructions below,
we will use the command "gmake" to emphasize the importance
of using GNU's version of make.

To install this code, do the following:

isip22_[1]: mkdir /u0/smith/
isip22_[1]: cd /u0/smith/
isip22_[1]: mv $HOME/snr.tar.gz .
isip22_[1]: gunzip snr.tar.gz
isip22_[1]: tar xvf snr.tar
isip22_[1]: cd snr
isip22_[1]: export ISIP_SNR=`pwd`
isip22_[1]: echo $ISIP_SNR
isip22_[1]: export ISIP_CPLUS_COMPILER="g++"
isip22_[1]: echo $ISIP_CPLUS_COMPILER
isip22_[1]: cd class/signal/v1.0/
isip22_[1]: gmake install
isip22_[1]: cd ../../..
isip22_[1]: cd util/snr_calculator/current/
isip22_[1]: gmake

Following steps can be used to test the code. Because of
different byte-orderings on different machines, we 
need to be careful about byte-swapping the data.

If you are working on a Sparc, do the following:

isip22_[1]: cd ../../test
isip22_[1]: ./test_sparc.sh

If you are on a x86 machine, do the following:

isip22_[1]: cd ../../test
isip22_[1]: ./test_x86.sh

You should get the following output:

File: ./710_b_8k_sparc.raw
         snr for channel 0: 16.034065 dB
File: ./710_s_8k_sparc.raw
         snr for channel 0: 33.657867 dB
File: ./711_g_8k_sparc.raw
         snr for channel 0: 7.642164 dB
File: ./712_f_8k_sparc.raw
         snr for channel 0: 12.993533 dB

(For x86 machines, "_sparc" will be replaced by "_x86".)

Additional information about the utility can be found in its help message:

isip22_[1]: ./snr_calculator -help
name: snr_calculator
synopsis: snr_calculator [options]
descr: produces the overall signal to noise ratio of the speech signal
example: snr_calculator -num_chans 2 -sf 8000 -sig_thresh .80 -noise_thresh 0.1
                        -frame_dur 10 -window_dur 20 -input input_file.raw

arguments:
        input_file.raw: 16-bit linear input file

options:
        -num_chans:    number of channels in the audio file (default: 1)
        -sf:           sample frequency of the data (default: 8000)
        -swap_bytes:   flag indicating whether the byte ordering should be 
                            swapped
        -sig_thresh:   percent of energy considered as signal+noise 
                            (default: 0.85)
        -noise_thresh: percent of energy considered as noise (default: 0.15)
        -frame_dur:    frame duration to use in computing the short duration
                            snr (default: 10 msec)
        -window_dur:   window duration to use in computing the short duration
                            snr (default: 20 msec)
        -use_hamming:  flag indicating whether or not to use a hamming window
        -use_pre_emph: flag indicating whether or not to pre-emphasize the data


-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
    Requests for support can be directed at help@isip.msstate.edu