Stereo Tool

Imagine what horrors employees have to endure when this process takes place in the vicinity.

_________________
Imaging producer & Stereo Tool audio processing consultant

My heart goes out to you. I'm across the world on Facebook Messenger, and it is already dangerous enough...

Found the cause of the Basstardizer issue. It was somehow mixing up the lookahead data of the two channels. Fixed now (next build).

Mwahahaha!!!! Well, you're at home and I'm still here at 2:55 AM. So... I still think it's worse for me

Update: Max Release Speed appears to work now.

Why release is a problem:

Attack
Let's say you're at absolute silence (sample value 0), the threshold is set to sample value 100, and some audio comes in at sample value 1000. Then the difference in level is a factor 10, and once we start moving, that decreases. Great.

Release
Now release. We have had a whole bunch of samples at 1000, and suddenly the input level drops to 0. The relative difference between 1000 and 0 is infinitely big... And even if we start moving towards it, the distance stays infinitely big. So, we keep moving up at an infinite speed.

Solutions
Option 1: Use max release speed. That will work in the next beta.

Option 2: Use a constant minimum level for sample values below that value (which could, but doesn't have to, be the threshold value). That would work, but it creates nonlinearity: The compressor moves the same regardless of whether we're at 2 dB attenuation or 200. That would no longer be the case with a constant offset.

Option 3: Use a moving minimum level for sample values, relative to the current attenuation level. I will add that asap, it will hopefully work slightly better (smoother) than option 1, which causes the speed to abruptly stop increasing above the max speed. Option 3 should have a much smoother transition area.

A new build with working option 1 is running now.

Don't attempt to run this preset until at least beta 30 comes out. However, it shows a working template for using relatively high values of max release, but enough to keep you out of trouble. Rather than worrying about a specific dB/sec, value. I focus on the numbers in the backend.

In multiband 1, I'm using 36 everywhere. That corresponds to a multiple of about 1.5 compared to the natural fall rate for a compressor. What I mean by that is that if you have a 2000ms release, then that has a natural fall rate of 6dB/sec (from 12dB of gain reduction). Using a 36 in the file for max release, would allow that compressor to release at a maximum rate of 9dB/sec. I went a little higher in the AGC, but not much. I use 38 (1.6 multiple) there.

I find these values are a nice compromise between letting the compressor be fully dynamic (and getting wild release), and restricting them to their natural rate, which robs you of some of the dynamic properties of symmetric mode.

Anyway, hope this proves helpful.

The build from last night was still broken (Wes found that before I even posted it here). New build has been running for a few hours now and should be ready in about 1-1.5 hours.

Stereo Tool 8.55 BETA030

NOTE: KNOWN ISSUES:
- Symmetric modes are visible for Digital compressor but is not (and will not be) implemented!

Windows 32 bit:
Windows stand alone: http://www.stereotool.com/download/ster ... 55-030.exe
Winamp DSP: http://www.stereotool.com/download/dsp_ ... 55-030.exe
VST: http://www.stereotool.com/download/vst_ ... 55-030.dll

Windows 64 bit:
Windows stand alone: http://www.stereotool.com/download/ster ... 55-030.exe
VST: http://www.stereotool.com/download/vst_ ... 55-030.dll


CHANGES:
- Max release speed fixed (*)
- Basstardizer: Fixed issue with difference between left and right
- Crash reported by near05 (fixed?)
- Improved external RDS insertion to allow for noisy RDS signals (some encoders (actually apparently from one specific brand) generate a pretty bad signal, which was rejected by Stereo Tool).
- Added sound card lowpass filter compensation to FM output sound card (so now beside tilt for bass there's also something for highs). NOT YET FINISHED, may cause weird behavior. It will work in many cases, but not in all. Does not work when "Ignore high frequencies" is used. Description will follow when finished.

(*) I have added some code to generate a pretty bizarre curve (which does give the correct behavior). I want to get rid of this code, the behavior could very slightly change when I manage to get rid of it.

OLDER CHANGES:
- New compressor mode. Check all weird settings, optimize, new peak mode, remove faster attack/release, add setting for max release, fix max attack/release speed controls. Things to check: "Peak", "Peak mode", "Feedback", "Feedback^2", "Exponential release", "Max attack speed", "Max release speed". New mode is called Symmetric RMS, there's also a Symmetric Averaging but I will probably remove it because it's not useful. MORE INFO BELOW!
New AGC mode MORE INFO BELOW!
New Auto EQ and Bass EQ mode (not switchable in Bass EQ since that's new anyway). MORE INFO BELOW!
- RDS: Add PTYN
- RDS: Add timeout for TA
- Improved multipath protection
- Improved Stereo Image phase limiter
- Revisit new Hard Limit / Final Limiter with Matt & Wes - Settings removed Waiting for final feedback.
- MPEG detection is too insensitive now -> Probably fixed Waiting for feedback
- UECP UDP mode Waiting for feedback
- TA/TP via MagicRDS. Auto-enable TP if TA is used? (no)
- Added THIMEORT in ASCII RDS mode.
- Spectrum displays are broken in 4K mode
- Compressor/limiter on input before AGC to handle extreme sounds? / 2nd AGC with higher threshold for voice tracks Added, may need new mode. Waiting for feedback.
- Add auto save to Linux build Waiting for feedback
- VST delay appears to be wrong - Phase Delay and Threads
- VST version no 4K
- Backup input - fix overshoot display
- Backup input - check stereo
- Fix web server long URL issue Test release mode needed
- Scrolling through presets in 4K mode locks screen (crashes).
- Pilot-only output for 9 , works but needs to be sent back to O9SG
- Self test crashes
- Fix affinity for Linux and Mac - Mac doesn't support it.
- Edwin's bass effect - is that done now? (Maybe an extra HPF?) -> Not needed for now, based on preliminary tests...
- Put Bass EQ under license
- Do we need multiple Basstardizer bands? -> Doesn't seem like it. Not for now.
- Remove Hans mode from Bass EQ? -> Hid it (Extreme Tweaker mode, Matt mode is default, made it a pulldown instead of a boolean to clarify its behavior).
- Octaviafm extra logging? - Waiting. Once we know the cause I'll run a new build. But I'm not going to hold off releasing the current version, since 1. this issue doesn't have any side effects anymore and 2. it very likely isn't new. -> Nothing heard anymore
- Would multiple where_i_am start points improve attack? YES!!!! Think about how to implement it...
- Enabling low latency output caused crashes
- NAN recovery added to Basstardizer
- Top area of GUI made a bit higher, not all menu items fit before
- Bigger font and widgets on 4K screens
- Added Basstardizer
- Remove warnings that were shown for things that weren't even enabled.
- Fixed NAN in 8.54 in several versions in the clipper (reported by 1 person, appears to have been settings-related).
- Matt's bass effect Ready to get feedback, see "BASS EQ" before or between multibands! See note below.
- Fixed Mac multiband compressor gate error after NAN recovery.
- Add Breakaway Pipeline warning
- Add low input level warning
- Add too little AGC action warning
- Add too much noise gate warning
- Add too much BS412 action warning
- Added ASIO latency setting.
- Made streaming with libVLC easier, WARNING though, the audio quality is still the VLC streaming audio quality, there are better encoders out there. This only affects the stand alone versions.

NOTE: Bass EQ
1. This thing should not really be needed because there's already Auto EQ. However Matt Levin insisted that it's useful, especially now that they are playing very old Christmas music, so let's try it out!
2. There are 2 modes: Matt Mode (see checkbox) or Hans mode (unchecked). In Matt mode, the highest band is not adjusted but it is used to determine the levels. So in the last stage, the highest band is in bypass mode, but the other bands respond as if it's working as well. So the bass is adjusted compared to the full spectrum. In Hans mode, the highest band is ignored completely, so if you use 2 bass bands, they are adjusted to match each other, but not the rest of the audio.

TODO:
- Symmetric modes are visible for Digital compressor but not implemented!
- MB2 band coupling bug: something is wrong, check better-fg-plus.sts -> Seems ok now. Weird.
- 4K mode line spacing bigger for presets
- Test NanoPi
- Classical preset
- NAN seen when switching to new compressor mode which also doesn't get recovered - silence afterwards. (Peak mode recovers it though).
- Make default window size bigger for 4K
- Optimize new compressor mode
- Make 9/9sg pilot generation smooth for external feeds for DARC etc.
- Bass mono fiter to reduce multipath
- Is pre AGC affecting audio levels? (With BS412) -> Not specifically for BS412, but it slightly reduces the level even if set a lot higher. Probably needs some offset.
- Aanvraag uren
- AM streaming output
- Basstardizer fadeout (Edwin)?
- Add AM presets
- Check MB2 1% something (see FB chat about crackling sound, which was ok in 8.00). Test with Rhodes54 preset -> Ok. But Spideys Web (7.41 vs 8.54) is different, let's test that.

- AF method B via UECP
- 18 Hz trigger tone
- Declipper check L-R effects
- Reduce extreme stereo
- Raspberry Pi "gedeelde bibliotheek" melding.
- Check out Butt
- Add libVLC instability warning
- Loudness meter (also for streaming)
- Omnia.9 set Stereo signal. Probably fixed? Wait for answer.
- Omnia.9 test/fix logging
- Omnia.9 AutoPilot add flag for generating noise above pilot?
- WatchCat website
- Triple stuff
- Thumb up/down in GUI for presets

- SCA2 use as backup, feed to output without processing, bypassing normal chain. Use to have FM receiver as backup, and then just feed the full signal to the output.
- uMPX 24x2ch interface
- One bass page

- GUI: Hide disabled stuff, far less confusing / much cleaner interface.
- Talk to chip
- 2 jaar
- Fardau
- -24 dB side chain range?


New compressor mode info:
About 3 years ago I wanted to redesign the whole compressor based on how water pressure control systems work. The whole idea was to have a compressor that moves fast when it has to, but stands still when it can. And that it would behave 'naturally' - it's something that you could actually build with real components (if you treat water pressure as audio levels).

This appeared to work pretty well. If you play a dynamic track, the compressor acted fast, and if you play an already compressed track it does nearly nothing. This is great - because it doesn't add 'restlessness' to already compressed audio, which would lead to listener fatigue.

Short summary of the rest of this text: It never worked as I envisioned! (And I never even realized that). But now it does!!

Some people have complained about the attack sometimes working too fast - which was mainly noticeable in loud 'S' sounds. The fast attack with much slower release could cause a 'gap' in the audio after the S. But I figured that that was a more or less logical consequence of the the whole act-fast-when-you-have-to idea.

Now fast-forward a bit. Last year, 2 people (Wes Keene and Matt Levin) were working on a preset together with very high compressor ratios and they noticed that in one particular track, a vocal which was originally louder than the surrounding sound actually came out softer. Well, that's not good. It sounds really annoying and broken. After looking into it I found out that my design was actually not - as I had thought - controlling the level based on RMS levels, but based on average sample levels. In most cases that's closer to RMS than peak, but it's not perfect, and as it turned out, in some cases flat out wrong.

So, I added a new mode that actually controls things based on RMS levels. And - different than many traditional compressors - it would control based on RMS levels regardless of whether it's long time or short time levels that you're looking at. Typically, a compressor would use some window over which it calculates the RMS level, and then control the audio based on that. That works great for most content, but if you play a track that for example has a very loud kick every second or so, the RMS window won't help a bit and your long-term level will probably be too low.

So I created a new RMS mode, renamed the old one to Average mode, and thought I was done. But, the problem with loud "S" sounds has gotten way worse! And it was even worse than that: Release was slower than before. So to fix that, I added some "patches": "Big release speedup" and "Big attack slowdown".

This new mode doesn't need that anymore. It's completely symmetrical, and it's still RMS-based, both on short and long time intervals! It still moves slowly when it can and fast when it has to - but not insanely fast if the attack is really big. And release isn't very slow anymore.

Since attack and release speeds are now virtually identical for RMS and Average mode, and you probably prefer RMS levels over Average levels, the new "Symmetric Average" mode is basically useless. I've put it in for now for testing, but I'll remove it afterwards.

The AGC and Auto EQ were suffering from similar issues, and I've fixed those too. There you can just disable "Legacy mode" - if I'm not wrong, the new mode should be better for any purpose than the old one.

Important: The CPU load is really high right now. I will optimize it later.

Also important: You can't just enable this new mode and expect everything to be fine without changing the settings. For small level differences there won't be a difference, but there will be for big differences. You'll definitely notice the slower attack on big level jumps.

What actually happens when selecting Symmetric RMS in Digital compressors in the last 2 betas?

That "fade in" issue at the beginning of tracks still seems to be present.