Stereo Tool

Changes in this version:
Behavior of Highs Priority improved further: Even less drops.

Tomorrow I'll improve the de-esser.

CPU load is still higher than in beta005, but should be a lot lower then in the last few beta's (same as 013).

Stand alone: http://www.stereotool.com/download/ster ... 32-014.exe
DSP: http://www.stereotool.com/download/dsp_ ... 32-014.exe
Other versions not available.

Older changes:
- Bass sensitivity clipping level drop: If there are both mids and highs, the drop is a bit less strong than before
- Bass sensitivity clipping level drop: Determining more precisely when the clip is needed. So instead of dropping the bass the whole 4096-samples block, it's now only dropped at the locations of this block where there really are mid/high freqs
- Bug reported by Phoenix ( viewtopic.php?f=15&t=4235 ) should be fixed.
- BS412 also needs a max deviation (+/- 75 kHz) display --> Added
- BS412 sometimes caused drops of upto 2.5 dB. This is preset dependent. --> Adding a slider with which you can specify the amount of headroom; by lowering the output level by 0.05 dB the margin for error is already increased by 50% which seems sufficient. Test if adjustment adjust compressors properly.
- Also adding a BS412 slider with which the max deviation can be lowered; this way you can create a denser sound (apparently many commercial stations like that even though they don't gain any loudness) and it protects against spikes when the sound card or path to the transmitter isn't perfect (reason: Many current processors usually clip the signal at around +/- 50 kHz instead of 75, so in many cases the path to the transmitter doesn't have to be perfect). (Partially done, to be added: BS412 must take clipping into account; clipping level adjustment for pilot/rds).
- Something is broken in latency 512.
- Different up speeds for louder end result
- Sound card adjustment for low freqs needed (that was on my todo list anyway, but I'm moving it to the top now).
- Slightly changed sensitivity behavior (les steep drops and rises of bass). Not sure if this is the best possible solution: Alternative might be to let the bass drops start and end INSIDE the area where loud bass is present, or do the same as now but limit the maximum change speed.
- Made tilt configuration a bit easier (less steep changes at the extremes)
- 2nd RC slider is saved now
- Added displays so you can see what is being sent out to the sound card (including calibrations)
- Changed clipper distortion masking filter, this version should have slightly less distortion than the previous one but almost the same amount of highs.
- Added 2 extra sliders to control the distortion masking. Increasing the 1st or decreasing the 2nd increases the amount of highs. The goal is to find a combination of values which allows the highs to pass without adding (too much) distortion.
2nd slider determines how fast something is determined to be a tone instead of hiss - for tones I don't allow more distortion than before. First slider multiplies the calculated effect (that is, it multiplies the allowed amount of distortion, which doesn't mean that it will actually be used or lead to louder highs).
- Found and fixed several bugs in the clipper (hole punching by loud highs should be reduced).
- Allowing some extra very brief distortion to reduce hole punching by loud highs.
- Improved behavior of Highs Priority to avoid volume drops.


Planned:
- Check sensitivity


Note: High frequency square waves are NOT good for testing. Use square for low freqs, normal for high (above approx. 100-200 Hz).

@Brian: The saving is part of the redesign. That's why - currently - I have to implement it for each slider separately. Which is indeed kinda stupid.

For FM the difference in sound is really big. Not talking about scopes, it's really big. I *have* also measured it and I see that in some cases the drop of low/mid frequencies due to loud highs has been reduced by upto 4 dB - that's quite a lot and very clearly noticeable. Also the highs sound more natural and open to me (again, determined by listening, not measuring). For non-FM the difference is probably a lot smaller (haven't checked it yet).

I'll try to optimize it further tomorrow, actually I hadn't expected such a big rise so there must be something there that uses more processing power than I expected.

@mapleleaf: Are you using ASIO with low latencies? From your feedback I still get the impression that this might be caused by something like a too low buffer size. (And that would indeed be fixed if you reduce the latency setting).

So as it is right now, there is not the capability of a parameter get on program load and a set on closedown?
I can tell you based on my ears that there's not a heck of a lot of difference for non-FM, if there is any difference at all. If it is above 15K though, I might be having trouble hearing the change.
If there isn't a significant change for non-FM, perhaps you could make it to where the settings are only enabled if FM processing is enabled.

Big Hans, beta 14 ..... I have no words to describe it .... well, very well


best

I now have that CPU load at a little less than 10% above that of version 6.31. The things that caused most of the increase have no measureable effect anymore, so they are optimized far enough. What remains is the code that checks how much drops occur (I had optimized that code too far in the past, and due to that it didn't function properly. Now it does but at a CPU load cost. Will check if I can do something about that).

Next version will also have a completely redesigned de-esser (when pre-emphasis is used), which has far less effect on the sound.

Edit: Slightly reduced the precision of the "S" hole estimation. I don't expect this to be noticeable at all - and for loud S sounds there's virtually no difference in output (0.01 dB). Now at 5% higher CPU load than 6.31.


I'm running a build now, will upload this version in about 20 minutes.

Changes in this version:
De-esser improved.
Some performance optimizations.

CPU load is about 5% higher than in the released 6.31 version - on my system, with all filters enabled. Difference is caused by the "Highs volume drop protection" (which is by default turned off for non-FM.)

I just realized that there may be another optimization possible, will check that later.

Stand alone: http://www.stereotool.com/download/ster ... 32-015.exe
DSP: http://www.stereotool.com/download/dsp_ ... 32-015.exe
Other versions not available.

Older changes:
- Bass sensitivity clipping level drop: If there are both mids and highs, the drop is a bit less strong than before
- Bass sensitivity clipping level drop: Determining more precisely when the clip is needed. So instead of dropping the bass the whole 4096-samples block, it's now only dropped at the locations of this block where there really are mid/high freqs
- Bug reported by Phoenix ( viewtopic.php?f=15&t=4235 ) should be fixed.
- BS412 also needs a max deviation (+/- 75 kHz) display --> Added
- BS412 sometimes caused drops of upto 2.5 dB. This is preset dependent. --> Adding a slider with which you can specify the amount of headroom; by lowering the output level by 0.05 dB the margin for error is already increased by 50% which seems sufficient. Test if adjustment adjust compressors properly.
- Also adding a BS412 slider with which the max deviation can be lowered; this way you can create a denser sound (apparently many commercial stations like that even though they don't gain any loudness) and it protects against spikes when the sound card or path to the transmitter isn't perfect (reason: Many current processors usually clip the signal at around +/- 50 kHz instead of 75, so in many cases the path to the transmitter doesn't have to be perfect). (Partially done, to be added: BS412 must take clipping into account; clipping level adjustment for pilot/rds).
- Something is broken in latency 512.
- Different up speeds for louder end result
- Sound card adjustment for low freqs needed (that was on my todo list anyway, but I'm moving it to the top now).
- Slightly changed sensitivity behavior (les steep drops and rises of bass). Not sure if this is the best possible solution: Alternative might be to let the bass drops start and end INSIDE the area where loud bass is present, or do the same as now but limit the maximum change speed.
- Made tilt configuration a bit easier (less steep changes at the extremes)
- 2nd RC slider is saved now
- Added displays so you can see what is being sent out to the sound card (including calibrations)
- Changed clipper distortion masking filter, this version should have slightly less distortion than the previous one but almost the same amount of highs.
- Added 2 extra sliders to control the distortion masking. Increasing the 1st or decreasing the 2nd increases the amount of highs. The goal is to find a combination of values which allows the highs to pass without adding (too much) distortion.
2nd slider determines how fast something is determined to be a tone instead of hiss - for tones I don't allow more distortion than before. First slider multiplies the calculated effect (that is, it multiplies the allowed amount of distortion, which doesn't mean that it will actually be used or lead to louder highs).
- Found and fixed several bugs in the clipper (hole punching by loud highs should be reduced).
- Allowing some extra very brief distortion to reduce hole punching by loud highs.
- Improved behavior of Highs Priority to avoid volume drops.
- Improved de-esser


Planned:
- Check sensitivity


Note: High frequency square waves are NOT good for testing. Use square for low freqs, normal for high (above approx. 100-200 Hz).

Checked that most of the CPU load comes from checkboxes "if preephasized" and "HQ" and when enable the Advanced Bass Distortion Protection ( ABDP ) . From there comes that jumps in CPU from 22-24 to 50% that i mentioned before.
Any checks on that Hans ? Or thats it for now.

The load is indeed back down to manageable levels. Like DJ-DOGGY mentioned, the bulk of the load increase for me is now due to using the Lossy Compression pre-emphasis while I am working on the Cobalt preset, which I'm intending for web radio. If I don't have that enabled, load is down around 40%, while it's 65-70% if enabled (causes the highs drop protection to kick in). The good news (for me) is that I would not normally have the lossy compression enabled for myself...

Hans

No I am not, I just set the latency to 2048...I can live with that

mapleleaf

Post 100 - it's getting time to release this version

A few updates coming up:
1. Reduced CPU load a bit further and at the same time made some changes that allow loud S sounds to be slightly louder (almost no drop effect - less than 0.1 dB, and sometimes even a reduction of the drop compared to the previous version).
2. Improved BS412 algorithm a bit (will have to test this to be sure) - it should stay a bit closer to 0 dB than before with nearly no effect on the audio. (maximum effect on the audio level is 0.3 dB so that should not be noticeable).