Stereo Tool

Since Beta 035, this meter no longer works correctly on my PC. I think it is just a GUI problem because I don't hear any difference. I tested ST on another device and the meter works correctly there.

We have a clipper improvement (hopefully) coming in tonight's beta.

There are 2 new things here:
- Highest vs lowest highs: stricter protection
- Reduce spikes
The first works on both L/R and composite mode, and makes clipper stages slightly stricter, which can slightly reduce volume effects on other frequencies, and on lower highs from loud higher highs.
The 2nd on, "Reduce spikes", for now only works on the composite clipper (it's for testing for now); if this works well we'll also make it work in L/R mode.

Reduce spikes needs a bit of an explanation. It may require some tweaking.
Say you have a very loud high frequency tone in your signal. Currently, that pushes everything else down:
(Cyan: Audio without that high pitched tone. Purple: Audio with that tone).
All frequencies are lower, the highest bit of the highs is about 9 dB down
The difference is also very visible in a spectral view:
With this new filter enabled, here's cyan without vs purple with again:
And the new spectrum:
This wasn't really a problem before, but recently we found out that setting "Highest vs lowest highs" a lot higher made the highs much better. But that does mean that we're pushing the highest bit of the highs (with the most pre-emphasis) very hard into the clipper, and without this new protection that causes problems.

Thanks to @Bojcha for complaining about this.
As you can see in the images above, this affects all frequencies, even the mids and lows. In this (really extreme) case the difference is about 1 dB.
Now, one thing: If you enable this, it tends to remove some of the "ringing" tones in S sounds etc which - at reasonable levels - actually sound nice. So, that's why there are strength and threshold parameters. With these 2 you can configure it such that small frequency spikes are still allowed to push things down a bit. This will likely need to be tweaked.
Finally, "Carve gap" can be used to make a gap around these loud tones. This may sound more natural, I haven't heard it yet, but there's some logic in that (preventing extremely steep filters). The effect looks somewhat like this:
This will be in tonight's new beta (042). Again, COMPOSITE ONLY for now for this 2nd filter.

This should normally have only a small impact on music, except if there are insane sounds in it, which really can happen. For example, here's half a second of a song that Bojcha sent me with a really really really weird "S" sound (I still don't know if I like it or hate it); cyan is output without this filter, purple is with this filter enabled:
I don't know how anyone manages to sing like this...

I've just been playing with settings a bit, and I think that the behavior is better if I raise "Reduce spikes higher than" a bit and raise the "Spike removal strength as well. At 2 dB and 400% (instead of 0.8 dB and 300%) I get the same protection for the weird S in this song and the high frequency tone, but less impact on other audio. But I'll leave it as-is for now to gather feedback.
The disadvantage is that more highs that push everything else down will be getting through. But maybe it's still low enough to not be a problem.
Since they are new and mostly for testing, both filters are disabled by default - so go to Advanced Clipper - Bass/Highs hole punch protection, and enable them from there. If they are really always better and we have good default values, we may enabled them by default in a future build.

A bit more info.
This is 10 minutes of music. Cyan is without the new protection, purple is with. So based on this, audio between about 6-13 kHz is reduced.
Now that seems bad. But if I zoom in on a few seconds of audio, and look at the difference signal (note: Difference, so anything here can be louder OR quieter), it looks like this:
As you can see, typically there are a lot of horizontal lines, indicating a single frequency that's reduced in the end result.
If I select one of these sections and look at the spectrum:
It's clearly not the whole highs section that's reduced, only some spikes in it. Which may still sound nice. I don't know.
And in many cases the stuff around it gets louder.

BETA042 is uploaded. Let us know what you hear!

Update: First feedback that I've received is that strength values > 100% are problematic. For > 100% we enter a different, more aggressive clipping method, which is destroying details in the highs. That probably doesn't mean that that mode should _never_ be used, but it needs to be restricted to the most extreme spikes.

Also, other reported issue: Just a loud tone gets pushed down at values > 100% if you drive the clipper really hard. That also needs to be prevented. (We will limit that mode to extreme spikes, but it still needs to not push it down).

After a lot of feedback and multiple new beta builds, I think it's working properly now. Except that some people have weird vibrating sounds, which are apparently CPU specific - so we need to figure out what's causing those.

BETA044 will have the improved behavior - but sometimes with those vibrations.

Edit: Vibrations cause found; fixed, and this new filter now also works in L/R mode.

BETA044 Posted

_________________
NB: I'm not a professional, just a hobbyist, and as such I could be wrong. If so, please, somebody correct me.

Could be user error on my part, but the Mac standalone version installer ThimeoStereoTool-Installer_BETA1076-044.dmg seems to install 034 instead of 044.

We keep getting errors back from Apple's signing servers. That might be causing this. We'll try to check what's going on asap.

Update: BETA045 adds 3 new things:

• The "aggressive" highs filter from the previous beta now has a fixed minimum threshold below which it never acts. This may avoid overreactions. Not configurable; it's probably unnoticeable (if the protection does less than before, then I need to fix it).
• That midhighs protection (500-5000 Hz). It may do too much and damage vocals. I have so far only had time to test it on tones on top of noise; on those it does exactly what we want.
• Bass protection fix. The mids and highs vs bass IMD protection has 2 "strength" sliders; if they are set high it can happen that the bass disappears completely, and it can even come back with inverted phase. Which would also cause distortion again. There's now a slider that controls the maximum amount of bass that can be removed. Setting it too low may cause IMD distortion

Image below: Spectrum without (top) and with (bottom) this new mids protection. Noise with a sweep. The gap that's seen on top is about 8 dB deeper than at the bottom.

I have a lot of changes coming up. I think the behavior is much better now.

• Internally, there was a 'minimum' value below which the aggressive mode would never activate. That seemed to make sense, and it does for pure constant tones. But no longer if the frequency moves around quickly, for example on a high pitch tone in an S sound.
• Threshold, specifically for highs, has been changed, it used to only look for very narrow spikes in the spectrum (less than 1000 Hz wide), that's now increased to about 6 kHz. Which might be too much. But before it would 'miss' high piched tones on S's that weren't a pure frequency, but just a relatively narrow spike. IT WILL CATCH MUCH MORE NOW at the same setting as before.

I have tested both changes on music and it looks like there's very little effect on normal music.

• The Aggressive Strength sliders both have been rewritten and work completely differently from before. They seem to be fully active now at 141%; I recommend settings them to that value (but: see below). (Bojcha has already been testing with this and he arrived at 108%, but again: see below)
• I have added a new "Strength #2" slider. Yes, great name. I think that that slider can replace the existing Strength slider, if you leave that at 141%. My guess is that "#2" works far more naturally. The default setting is 6 dB (which is what it was before as well, there just wasn't a slider for it). If I lower it to 5 dB, and play pink noise with a loud 13 kHz tone, the 13 kHz tone comes out 0.8 dB louder than before, while the noise only drops 0.2 dB. Measured output level (even after demodulation) is also higher if I use this.

With Strength at 141%, Strength #2 at 5 dB, the sweep image from above now looks even better: