Stereo Tool

Here's another one: Bjork - It's oh so quiet.

In this track all the artifacts are caused by Final Limiter or Loudness. At 1024 they are gone! (which is kinda surprising).


EDIT: FOUND IT! IN 512 SAMPLE LATENCY MODE, TURN 'STRICT BASS FILTERING' IN LOUDNESS OFF! I'll do this automatically in the next version.

And, if you don't get enough bass, adjust Multiband (I know, this should also be done automatically).

Phase rotation also causes very bad artifacts (for acceptable levels set it to: initial speed 5000, rotations 3, smooth 99.8%, start freq at least 480.). Of course it's far less effective this way.

Also, make sure to get rid of sounds below ~ 30 Hz (set highpass filter to 30 Hz, not phase linear!)


So to get good audio at latency 512:
- Highpass at 30+ Hz, not phase linear
- Phase rotation OFF or VERY VERY smooth (480, 5000, 3, 99.8%)
- Loudness: Turn 'Strict bass filtering' OFF.
- Don't use clipping for the lowest 3 multiband bands, or you'll get 'distorted' bass sounds (like a subwoofer that's overdriven). There's no way to solve that at such low latencies. Or at least use very little of it.

The first 2 probably also improve the 1024 sample latency sound - which is quite a lot better than the 512 sample latency. (The Loudness thing doesn't seem valid there though, and Highpass can be set a bit lower).

Bojcha: Where do you find those tracks?!

(Actually I love it - just like the Beyonce/Gaga 'Telephone' track you mentioned earlier).

I've listened to it with your preset, with the changes I described in my previous post, and I don't really hear anything bad. Yes, the quality is less good than that of full latency (which makes sense), but the difference seems pretty small... Or am I missing something?

There are planty of those track with "nasty" bass.

I tryed your settings for "512" mode.
It's bad that "Strict bass clip" must be off, but anyway i don't think that "Strict" is problem for 512 mode.
Still think that multiband makes major problem. Mutliband is not good even in 4096 mode because same reason.
40Hz band and 90Hz band are not 100% separate. so tones at 90Hz will have an impact at 40Hz and 170Hz band too.
I know that in 512 mode those bans are "joined" but that's bad anyway.

test2:
Input: 90Hz Sine tone, Any preset
Something happens with AGC when you switching 4096 and 512 mode. Like in 512 mode input is higher and AGC goes down.

That's possible. The AGC is based on RMS volume, but the bass is counted far less strong.
In 512 sample mode, the bass filtering (which is needed for the input of the RMS level calculation) is also less precise than at higher frequencies.

Assuming 44.1 kHz input, if you use 4096 samples, one block of audio (for which the RMS level is calculated) contains 4096 / 44100 * 90 = 8.36 complete waveforms.
At 512 samples, that's 512 / 44100 * 90 = 1.04 - not much!

(In fact, anything below 86 Hz does not fit in 512 samples, that's why frequencies below 86 Hz give the most issues; frequencies above twice that (172 Hz) are almost problem-free. For 4096 samples these values are 10.8 and 21.5 Hz, on top of that the artifacts that are caused are 8 times softer).

I've been Googeling a bit and I now at least know the cause of the artifacts at lower latencies.

Well I did already know it, but now I know what it's called: time aliasing. Basically, if I have a short piece of audio and filter it, the audio can 'spread out' a bit. During processing I treat a single block of audio as if it repeats infinitely. So if audio spreads out, it can move into the next or previous repetition, hence 'time aliasing'.

The solution is simple: Add some silence around each block of audio, and then also merge the audio in that silence block back into the end result.

Unfortunately, that would at least double the CPU load, and I'm still thinking what effect it would have on latency (probably none, as I need everything outside the audio window to be '0' after Final Limiter/Loudness anyway).

I also found one solution to reduce the effect as much as possible - but it turns out that that was exactly what I was already doing........

This is interesting: I've made some changes, and I clearly did some things wrong (if I play a deep bass sound, I get horrible artifacts).

But normal music sounds MUCH better! So I think I am looking in the right direction.


Edit: With new filtering method, highpass filter at 35 Hz:
- AGC: OK, no artifacts
- Bass boost: OK, no artifacts
- Loudness: OK, no artifacts (!)
- Stereo: OK, no artifacts
- Noise gate: OK, no aftifacts.

But:
- Multiband: Bad, lots of artifacts, especially when Clipping is enabled.
- Phase rotation: Just horrible.

Hans great news! I still have to upgrade our station to the latest stereotool version (we at 5.10 still)

Btw, I was reading some article on a band where the sound engineer messed up one of their albums while it was being produced, apparently they did not filter out the frequencies below 40hz and it had the possibility of making the high freqs inaudible.

Not that great yet - turns out I'm trading one type of artifacts against another. So now I'm trying to find something like a 'sweet spot'...

I'm now building a new version.

In the end, I only changed one value by 1 %. Still, the effect on certain sounds is very noticeable.

You can expect the following changes:
- Loudness: No more 'vibration' in highs when loud lows are present (not the same as voice vibrations, it's something much slower and very annoying, only present at latency 512). Very clear in 'Bjork - It's oh so quiet', first few seconds.
- A bit more overall distortion in Loudness in return, mainly for loud lows. Also noticeable in those same first seconds of that Bjork track. Apparently, and I don't understand this yet, it disappears when you increase 'Increase dynamics' in Loudness (which I expected to increase the amount of distortion - this probably means that I'm doing something wrong in my filtering somewhere...)

There's no need to turn 'Strict bass clipping' off anymore - that issue is solved now.

Phase rotation is still horrible - so turn it off or set it at as safe as possible values (>480 Hz, 5000 Hz, 3 rotations, 99.8%).


So, for good sound:
- Highpass filter to Automatic selection or Never phase linear
- Phase Rotation to 700 Hz, 5000 Hz, 3 rotations, 99.8%
- Loudness: Increase dynamics to +15% (or higher) - ONLY needed for 512 (which probably indicates a bug in the latency 512 filtering)

Building now!

Winamp DSP plugin: http://www.stereotool.com/download/dsp_ ... 01-001.exe
Stand alone version: http://www.stereotool.com/download/ster ... 01-001.exe
VST version: http://www.stereotool.com/download/vst_ ... 01-001.dll
VST version (No SSE2): http://www.stereotool.com/download/vst_ ... 01-001.dll
Command line version: http://www.stereotool.com/download/ster ... 01-001.exe
Linux command line version: http://www.stereotool.com/download/ster ... ETA601-001 [not available]
Linux GUI version: http://www.stereotool.com/download/ster ... ETA601-001 [not available]

WARNING: This is just a test - I'm not saying that this version is better than version 6.00. And only the 512 latency sound is affected.

At latency 512, for good results, set:
- Highpass filter to Automatic selection or Never phase linear
- Phase Rotation either OFF or (around) 700 Hz, 5000 Hz, 3 rotations, 99.8%
- Loudness: 'Increase dynamics' to +15% (or higher) - ONLY needed for 512 (which probably indicates a bug in the latency 512 filtering). Setting it very high gives MORE artifacts at higher latency settings!
- Don't use clipping for the lowest 3 multiband bands, or you'll get 'distorted' bass sounds (like a subwoofer that's overdriven). There's no way to solve that at such low latencies. Or at least use very little of it.