Further to the files I posted a while back of calibrated S/N ratios.
and voice signal with poor S/N, I've now gone back and verified the
software and levels, double checking them against each other

The file      http://www.g4jnt.com/DownLoad/CalibratedNoise.zip
contains a complete archive of results.

In there are .WAV files of tones and a CW message at various Signal to
Noise ratios - all using mathematically generated tones and noise to
ensure accurate levels.

There are also several voice files at various S/N ratios.   The
reference voice file was recorded from BBC Radio 4 a while ago, then
filtered to the band 300 - 3300Hz to simulate the output from an SSB
filter.  The mean amplitude over the recording was measured and
various levels of noise added to generate the files.

The voice S/N is not defined in the same way as I did the previous
ones, so can't be directly compared.
Here I took the amplitude to be the mean rather than just trying to
estimate the phoneme peaks by measurement on a spectrogram

The spreadsheet in the .ZIP shows the parameters applied to all file
and how the values were derived.

Also included are a number of programmes files for generating .WAV
files of  single tone and  a CW message, and for adding noise to the
file at a precise level. and bandpass filtering a .WAV file

They were written using a rather archaic 16 bit prog language
(POwerbasic) and will run from a command prompt or by clicking on tthe
.EXE, but is not the fastest DSP software in existance.    It was far
easier to do the code in this language that I've used for years and
had all the routines and progs ready than to port to a more modern
language.
Source code is there if you want to check out

Andy
www.g4jnt.com