The fundamental frequencies of the human voice are well below the transmitted 
passband of an SSB transmitter.  However, the human voice has all harmonics - 
not just even or odd.  Therefore, if the algorithm assumes that the fundamental 
pitch is equal to the spacing between tones, it will correctly identify the 
fundamental pitch.

Once the fundamental pitch is identified, one needs to determine which of the 
present tones is the "lowest transmitted".  For a male voice with fundamental 
at 90Hz, for example, would the lowest transmitted pitch be 270 or 360Hz?  

The spectrum of the voice will tend to be stronger in odd or even harmonics, so 
a further subdivision may be performed, and this can help delineate whether the 
lowest received pitch is 2nd, 3rd or 4th harmonic.

The human voice has a declining energy spectrum of 5dB per octave, and this is 
surprisingly uniform from person to person and across the gender line.  
However, the use of EQs, "SSB microphones" and the like, renders this 
characteristic null on the airwaves, so it probably cannot be used to advantage 
in an algorithm.

Finally, I have not observed speech processing to alter any of this.  I have 
been experimenting the past few years with some mighty aggressive RF clipping, 
and the IMD created does not change the frequencies in the spectrum, only their 
relative levels.  IMD generates new tones, but mostly they fall directly on top 
of tones that already exist.  I've created up to 40dB of RF clipping in 
hardware (a task which requires one start with a very clean SSB signal to begin 
with) and my FFT-based spectrum analyzer shows distinct alterations in the 
slope of the response, and more noise filling in the blanks (particularly on 
sibilants which have virtually no frequency data in them), but the individual 
overtones are clearly present.

Dave W8NF

--- In [email protected], Leif Asbrink <l...@...> wrote:
>
> Hi Simon and Johan,
> 
> I have read an article on the subject. Presumably it was the QEX
> article you refer to, but I am not sure. I also spent quite some
> time in trying to implement an algorithm for SSB fine tuning
> into Linrad. That was not sucessful. Check with a waterfall display
> what the spectrum looks like. Stations that do not use speech
> processors have typically voice waveforms that repeat for a while
> and generate equidistant spectral peaks. It is not difficult
> to guess their common origin since the separation varies a little.
> 
> When heavy speech processing is used, a method based on power
> spectra fails completely. I can hear by ear so it must be possible
> to do it in the computer. I have no idea what kind of processing
> one would have to apply.
> 
> 73
> 
> Leif / SM5BSZ
> 
> > I don't have the QEX article - I think there was a bit more to it than that.
> > 
> > I don't have time at the moment but later this summer I will.
> > 
> > Simon Brown, HB9DRV
> > http://sdr-radio.com
> > 
> > 
> > > -----Original Message-----
> > > From: [email protected] [mailto:[email protected]] On
> > > Behalf Of Johan H. Bodin
> > > 
> > > I don't know how existing SSB auto tuning systems work but I think it
> > > could be done by looking for fundamental tones F and their harmonics
> > > N*F
> > > in the speech spectrum and adjust the tuning until they line up
> > > properly. Just an idea...
> > > 
> > 
> > 
> > 
> > 
> > ------------------------------------
> > 
> > Yahoo! Groups Links
> > 
> > 
> > 
> 
> --
>


Reply via email to