On Tue, 2007-08-14 at 12:10 -0700, Amy Alexander wrote:
> very cool! i've posted a comment but don't see it appear on the blog. does
> this blog wait for moderation or did it just send my comment straight
> to the /dev/null page?
They're held for moderation... Sorry I didn't realise that the software
didn't make this clear to the commenter, I'm going to switch to
wordpress soon I think.
> this is really cool! can you explain a bit more for relative laymen how it
> works? it might seem to the uninitiated at first to be a speech synth +
> vocoder, but it sounds like there's something else going on.
> in any case, i think it's an especially apropos + entertaining use of the
> visual side of livecoding.
Well it is like a really broken speech synth.
It seems in this area there's an important balance to be struck. On one
hand you don't want to necessarily make music with a speech synth,
because it's too much like a human voice. It's like it's difficult to
stop yourself trying to search for meaning and listen to the sound.
On the other hand I want the ease of composing sounds with text, where I
can easily play around with words, having some idea of what a word will
sound like. Also the results are a bit like speech, so hopefully a
listener can quickly get used to how the sounds are constructed and
relate to each other, because these relationships are similar to human
words/speech.
I guess it's also a bit like the idea of an 'uncanny valley' in
robotics. Broken speech synthesis sounds nice, but if it's more like
human speech, it just sounds rubbish in comparison, or maybe even
menacing.
On the technical side, the Karplus-Strong algorithm is just a delay loop
with a filter, with feedback. You put some white noise in the delay
loop, it feeds back on itself but because of the filter quickly smooths
out (rather than making the usual nasty feedback whistles). This acts
and sounds much like a real plucked string does, which is why it's
called physical modelling synthesis.
Once I made that I took two parameters; the length of the delay loop and
the probability 'blend' value that controls how 'drumlike' it sounds. I
picked a pair of values for these parameters for each consonant in the
English alphabet, my aim being to find a good range of sounds that sound
a bit like the letters I'm assigning them to, so fricatives harsher and
more percussive than more open sounds.
For the vowels I'm just applying a formant filter which really does make
it sound like human vowels.
I think what makes interesting sounds though are 'articulations'. I'm
not switching between the parameter values, but moving between them
quickly, creating diphthong type effects.
alex
Received on Tue Aug 14 2007 - 22:25:55 BST