BLuRry

After 11.5 years, I finally decided to switch jobs. Now I might have
some freetime to spend with my children and on hobbies again. (yay!)

Some side projects I wanted to get around to:

-Finish JACE (ok, not going to happen until Java has real joystick
support. It kind-of works, is that ok?)

-Apple // Sound Compression hacks (more on this, scroll down)

- 8-bit megademo (Using video and sound compression hacks and a 1mb
memory card to buffer the datastream)

Ok, so about the sound compression. I had this crazy idea a few
months back and emailed a former professor to gauge the feasibility of
the underlying theory. I think that I might be able to train a neural
network (using a genetic algorithm for tuning) to "hear" a small sound
sample and output the nearest approximate register values to reproduce
that snippet of sound using the 3-voice synth chip on a mockingboard.
The advantage of this approach over a table-driven one: Using a psycho-
accoustic perceptive model might yield more natural sounding results
with a higher level of compression than PWM with fewer compression
artifacts than low-bandwidth mp3. At least that's the theory of it
all anyway.

Training the neural net wouldn't be too hard, since I can just
enumerate through a random set of actual parameters and feed them
through an emulation to get the same snippet of audio from an
emulation. Of course, to get a real accurate training, I'll have to
collect the same sample set from a real mockingboard but I won't worry
about that unless I can prove it out in emulation first.

As for megademo, well it is not a shock that our friendly ][ is too
slow to blit full-screen animated hi-res graphics -- not in any sort
way other than limited-size scrollers. I might pull it off if my name
was Nasir, but that's not gonna happen with my skillset. But it IS
possible to pipe data through lo-res (and possibly double-lo) graphics
modes fast enough to achieve the necessary POV to make decent
animation. Since I'd already written the compression routines for
apple game server (and I did some preliminary testing in emulation for
video), I would only have to change the playback loop to read bytes
from a slinky card instead of from the serial port. Easy! But I kept
waiting until I did the sound stuff before trying this... and then
months and years passed by quite literally (ugh!)

Anyway, hopefully I'll have some progress in this regard. And if I
don't get very far, I'm more than happy to share code snippets and
ideas with anyone eager to reproduce these experiments for their own
self amusement. ;-)

-Brendan

Marc S. Ressl

On Nov 17, 8:02*pm, BLuRry <brendan.rob...@gmail.com> wrote:
> After 11.5 years, I finally decided to switch jobs. *Now I might have
> some freetime to spend with my children and on hobbies again. *(yay!)
>
> Some side projects I wanted to get around to:
>
> -Finish JACE (ok, not going to happen until Java has real joystick
> support. *It kind-of works, is that ok?)
>
> -Apple // Sound Compression hacks (more on this, scroll down)
>
> - 8-bit megademo (Using video and sound compression hacks and a 1mb
> memory card to buffer the datastream)
>
> Ok, so about the sound compression. *I had this crazy idea a few
> months back and emailed a former professor to gauge the feasibility of
> the underlying theory. *I think that I might be able to train a neural
> network (using a genetic algorithm for tuning) to "hear" a small sound
> sample and output the nearest approximate register values to reproduce
> that snippet of sound using the 3-voice synth chip on a mockingboard.
> The advantage of this approach over a table-driven one: Using a psycho-
> accoustic perceptive model might yield more natural sounding results
> with a higher level of compression than PWM with fewer compression
> artifacts than low-bandwidth mp3. *At least that's the theory of it
> all anyway.
>
> Training the neural net wouldn't be too hard, since I can just
> enumerate through a random set of actual parameters and feed them
> through an emulation to get the same snippet of audio from an
> emulation. *Of course, to get a real accurate training, I'll have to
> collect the same sample set from a real mockingboard but I won't worry
> about that unless I can prove it out in emulation first.
>
> As for megademo, well it is not a shock that our friendly ][ is too
> slow to blit full-screen animated hi-res graphics -- not in any sort
> way other than limited-size scrollers. *I might pull it off if my name
> was Nasir, but that's not gonna happen with my skillset. *But it IS
> possible to pipe data through lo-res (and possibly double-lo) graphics
> modes fast enough to achieve the necessary POV to make decent
> animation. *Since I'd already written the compression routines for
> apple game server (and I did some preliminary testing in emulation for
> video), I would only have to change the playback loop to read bytes
> from a slinky card instead of from the serial port. *Easy! *But I kept
> waiting until I did the sound stuff before trying this... and then
> months and years passed by quite literally *(ugh!)
>
> Anyway, hopefully I'll have some progress in this regard. *And if I
> don't get very far, I'm more than happy to share code snippets and
> ideas with anyone eager to reproduce these experiments for their own
> self amusement. ;-)
>
> -Brendan

One slave less, at least for now! :-)

I'm also glad to read all that encouragement you transmitted on csa2!
Thanks :-).

Marc.-