Re: Technical Question (was Digital Talking Book Standard )

[Nicolas Pitre <nico@cam.org>]

On Tue, 20 Nov 2001, Martin G. McCormick wrote:

> 	Yes, I am a bit slow, but I am catching on.  I definitely
> need to understand more about time scale shifting methods and how
> to accomplish them without adding distortion.
> 
> 	Each digit represents a moment in time and we can make
> things appear to speed up or slow down by intelligently inserting
> or deleting  information.

Exactly.

> 	If we do it on a sample by sample basis, we can make the
> recording appear to speed up or slow down with the expected pitch
> changes.  If we do it on a wave form by wave form basis, we can
> appear to keep the same pitch, but speed up the tempo or, for
> that matter, we can add extra wave forms and stretch out the
> syllables or whatever and slow them down.

Right.  And therefore you can accurately find out where the waveform 
boundaries are when dealing with digital sound.  The technique consist of 
finding the best correlation between the original signal with a small 
moving window 
of the same signal inside a limitted range.  You then get the exact sample 
position where the current waveform is likely to start and end.  Then you 
only need to duplicate or remove that waveform once in a while with a 
certain ratio to create the desired effect.

> 	The old pitch correcting devices like the one I presently
> use to read magazines butcher the sound because they aren't smart
> enough to make sure the next snippet of sound starts at the same
> place on the wave form that the previous one ended so we get that
> characteristic gravelly sound at high rates.

That's because those devices just don't care about signal periods at all, 
and tend to duplicate or remove an arbitrary fixed duration of signal.  


Nicolas