Stereo Tool

Been following the progress...

Was wondering if anyone else was having strange behaviour when adjusting the EQ bands in the MB using the last beta on Win 7?
Taking the slider all the way down seems to drop out the audio in that band / freeze up the metering.

Any chance of some predetermined frequencies (like Bessel Null) in a dropdown when using the test tone option for FM? I find it difficult to adjust accurately at present.

Well, Bojcha is making up for missing it by posting the shot of things not being the same height...

On a more serious note, is there anything at all further that can be done to optimize loudness (aka "clipper")? I'm barely able to use that now, even without pre-emphasis.

To give you an idea, here are some typical loads:

- Loudness on, pre-emphasis on w/ gap protection, declipper off -> 85-99% (GUI locks frequently, audio not stuttering)
- Loudness on, pre-emphasis off, declipper off -> 70-80% (GUI locks sometimes, audio not stuttering)
- Loudness off, declipper ON -> 40-50%
- Loudness off, declipper OFF -> 30-40%

Takeaway is that loudness in and of itself bumps the load up by 40-50% on an absolute basis. Stated differently, enabling loudness more than doubles the load of presets not using loudness when compared relatively.

My guess is that anyone using anything older than Core2 will see roughly similar results, meaning all P4, all K8-based Athlons, and all of the first gen Core processors (usually laptops).

Can you measure ONLY loudness? With and without oversampling? The jump that you describe is MUCH bigger than what I would expect, based on my own measurements the CPU usage of the declipper, multiband and loudness are very similar.

Edit: Hm no, that's not true. Both Loudness and the clipper are very similar, MB uses only one core so it's using less. (I measure maximum throughput for each filter, so for Loudness and Declipper I need to multiply by 2).

"Solved" , also solved a redraw issue that was introduced in the last beta.

I can measure only loudness, but I'm at a point where I don't understand the purpose of doing so. More than likely what will happen is the load will not be very substantial because disabling everything else allows loudness to utilize more of the data and execution caches.

I think a more "typical" usage pattern is what should be tested, not just a single item in isolation from everything else, but after I go get some bread and get a haircut, I'll give you those measurements.

That's exactly what I want to know. It's very well possible that disabling some other filter (if you're talking about cache/memory usage, Multiband uses about 5 times as much as Loudness) will also reduce the CPU load, so I want to know if it's really loudness that's causing it or the combination of everything - in which case focussing specifically on Loudness may be less useful.

There are some rather interesting optimizations that I can do on a number of other filters that can remove a number of - with some code changes unnecessary - memory copy operations (MB (40), AGC (6), noise gate (14) - numbers are the number of unnecessary copies PER SAMPLE). Both the declipper and Loudness (the last 2 filters that I wrote from scratch) are already optimized this way - declipper wasn't completely optimized until a few beta's ago though. Don't expect too much performance gain though, the change in the declipper removed 10 copies per sample and I'm unable to measure any difference from that...

That's indeed a bug - not only does the metering of the channel stop but the audio never comes back. In the previous GUI it wasn't possible to set these values to 0 - and it shouldn't be; this is probably causing some division-by-0 like errors from which the software cannot recover. Will fix it. Edit: Fixed
http://www.fmsystems-inc.com/index.cfm? ... ail&pid=56
So, if I understand things correctly, there are 5 frequencies at which the FM carrier frequency drops to 0?? Do I understand correctly that I need the 5 values in the last column? Also, I don't find the same numbers everywhere - do you know more about this? Setting these frequencies is extremely simple so that would be relatively easy to build it....

viewtopic.php?f=14&t=2661&p=7389&hilit= ... null#p7389

And while you there ...
Pink noise, and white one ..

It's both a combination of everything, and some things specific to loudness.

If I just have loudness on, no pre-emphasis or oversampling, then the loads are in the 30s. I can toggle the use shortcuts options and see no difference either.

Moving on to pre-emphasis and oversampling, if I use one or the other, the load stays in the 30s, but if I enable them both, the load goes up to the 60s.

If I leave pre-emphasis and oversampling off, but start adding back things one by one, I eventually hit a cache wall and it goes back up to the 60s again. If I push further, by enabling pre-emphasis, I get back to the 80s and 90s, like what I posted earlier.

So, there are various combinations that push a 1MB cache system over the edge.
I'd say that if you know that you can remove things without changing the audio, then anything removed would give more room for other data / instructions. Based on my experimentation, I'd agree that MB, AGC, and Noise Gate are candidates, but possibly Stereo as well.

BTW, I picked a nice long track to do this testing with... Pink Floyd - Echoes 23 minutes, 28 seconds