[Maybe a new RE] Opinion needed on dither algorithm, ...

[avasopht]

I've just been experimenting with a simple dither algorithm that I've had in my head for many years.

http://phead.mu/s/7UJa0WG4

To really test the algorithm I decided to do a 4-bit dither (if you look at the wave you'll see I'm really pushing it). The first 6 seconds or so is just pure truncation, everything after that is me tweaking parameters.

I'd like your second opinions on which portion of the audio features the best settings.

FYI, this isn't just adding noise. Noise is one of the tools, but the algorithm allows much more detail to pass through.

[normen]

Hard to say with such examples. Btw I suggest looking into what Paul Frindle said about dithering on gearslutz for example, he revealed a lot of information about the research they have done for the Sony Oxford console. Bottom line is you don't want your algorithm to put out too cyclic noise because the brain seems to pick up on patterns in the noise and do funny things with the perceived stereo width of the audio.

[avasopht]

Interesting read!

My algorithm is pushing all of the noise into the higher frequencies, though there a few ways I can spread the error. I've found that when you apply a low pass after this dither that it sounds comparatively clear (this was at 8 bits).

I've uploaded a second demo ( http://phead.mu/s/Xxi8XDIu ) with the following format:
Bar 1: truncated
Bar 2-3: increasing dither effect
Bar 4: dither plus low pass afterwards
Bar 5: original signal
Bar 6: dither plus low pass afterwards (to compare with signal)
Bar 7-8: like bars 5-6

When you view it through a spectrometer you can see that it's really just pushing the error into the higher frequencies.

[selig]

Is this not just "noise shaping", or something different going on here?

[avasopht]

Is this not just "noise shaping", or something different going on here?

You're right. See the thing with DSP is that there are a lot of names for the same thing, and because I'm more of an experimental person I tend to "come up" with solutions that already exist :frown:

After just reading the Wikipedia entry on Noise Shaping, seems like I was reinventing the wheel completely.

Well it was a fun exercise, at least I got my excitement for the day.

There is one thing I do slightly differently and that's amplifying the fed back error with noise (multiplying between -3dB and +3dB gain). Seems to let a little more detail pass through.

[selig]

Is this not just "noise shaping", or something different going on here?

You're right. See the thing with DSP is that there are a lot of names for the same thing, and because I'm more of an experimental person I tend to "come up" with solutions that already exist :frown:

After just reading the Wikipedia entry on

Noise Shaping

, seems like I was reinventing the wheel completely.

Well it was a fun exercise, at least I got my excitement for the day.

There is one thing I do slightly differently and that's amplifying the fed back error with noise (multiplying between -3dB and +3dB gain). Seems to let a little more detail pass through.

I get excited all the time with stuff like this, only to find out it's either already begin done or there are practical reasons why it's not! But believe me, it's far worse when it's a melody you think you've 'discovered'…

[Theo.M]

is it really possible to beat the izotope one anyway? Unless it was a very cheap RE i'd think that for $49 most would buy ozone to get the limiter and comprehensive megabit max dithering included which can work independently of one another anyway. Heck you can even put the ozone limiter to just deal with isps and nothing else, or just dither, or use everything but dither, etc.

[orthodox]

There were several discussions at Hydrogenaudio about which dithering algorithms are the best. No definite conclusion was made, but after reading those I decided to use pow-r3 or izotope mbit+ for production.