[DECtalk] DecTalk newbie

Tony Baechler tony at baechler.net
Tue Apr 19 22:34:36 EDT 2005


Hi.  I would like to expand on this a little, since I was messing around 
with Jayson's Daisy text file.  Actually, there are two serious problems 
with saving wave files from the demo.  First, 11 KHZ is the max sampling 
rate.  As most people know, that is fairly low quality nowadays.  I would 
say that 22 KHZ is a minimum for decent quality, but of course the only way 
to get that is by changing the sampling rate which causes problems as 
Jayson describes below.  The other problem is clipping.  That is actually 
what causes the distortion.  The best way to get around this is to save the 
sample at a lower volume, but I don't think the demo will let you do 
that.  The next best way is to record the speech with something like Sound 
Forge, RecAll, the Windows sound recorder or similar.  The best thing, of 
course, is to record directly from the hardware synth because clipping is 
minimal or nonexistant, but that is more complicated because you need a 
patch cord etc.  So no, there is no way to get it to harmonize with itself, 
but it isn't that hard if you have decent sound editing software.  Also I 
think there is some natural distortion that is part of the newer 
software.  My DEC Express uses the 4.2CD firmware and doesn't have 
distortion while the software Access 32 versions do, no matter 
what.  Actually I have noticed distortion with most software speech 
synthesizers.

At 07:32 PM 4/19/2005 -0400, you wrote:
>As far as I know, there is absolutely no way to make DECtalk sing in harmony
>with itself without using a wave editor.  DECtalk can only generate one
>voice at any given instant in time.  There is an option in the DECtalk demo
>to save a .wav file of the sound generated by the synthesizer.  Using that
>and a sound editor, you could put several of those files together to make a
>production, and that's what many people do.  Some, naturally, are of better
>quality than others.  Some files have distortion because, presumably, the
>voices weren't normalized before being combined, and/or the final product
>wasn't normalized.  Some have aliasing.  Aliasing is a high-frequency
>element added to a sound, where none was present before.  Aliasing is caused
>by people converting the DECtalk wav files to a higher sampling rate using
>certain poorly written tools for this purpose.  On the other hand, some are
>of probably the best quality possible, given the limits of the DECtalk
>system itself.





More information about the Dectalk mailing list