Stereo Tool

hi,
This is for overlap >0 only or for overlap=0 too ?

Edit: i see, it's for (=0)

Both.

Here's a copy of the todo list.

TODO:
- Remove (small) remaining peaks in Final Limiter - they are a bit higher now due to the improved bass handling ■ OK
- Fix very bad Final Limiter quality in low latency settings. ■ OK, fixed. Equal to v4.22 now, but I think I have some more improvements that I can try...
- Beep sounds are not completely constant in volume when they come out - check why and try to solve it. ■ OK, was caused by Pre Limiter, also Highpass filter
- Fix Stereo Boost - I found that it's behaving very differently from what I expected, but I don't know why. ■ OK
- Bass protection in Overlap=0 mode is less good than in Overlap>0 mode. ■ Nearly fixed, todo: add a selection checkbox to choose whether you want best sound quality or no very low frequency sounds
- Fix Singleband compressor. ■ OK
- Attempt to improve the highpass filter in Overlap=0 mode. ■ PARTIAL, works for 30 Hz phase linear at higher latency values (2048, 4096), there's now also a non-phase linear version for 20, 30 and 40 Hz that causes NO artifacts at all. Question: Should I ALSO offer the possibility to use v4.22 filtering? It's better at high freqs.
- Fix Hard Limit in Overlap>0 mode (completely destroys the sound at lower latencies at the moment)
- Attempt to reduce CPU load by simplifying processing: Dualband simpler implementation (4%) ■ DONE, Stereo Boost simpler implementation (7%), noisegate with less bands? (8%), combine Final Limiter with Loudness? (8%) - percentages is what these filters use NOW of the total CPU load in FM Loud Bass Europe mode - possible performance gain is a lot lower.
- Theoretically the CPU load of BETA3 should be clearly LOWER than v4.22 (a lot of optimizations should reduce the CPU load) - but it's only 2% lower. Check why. (Answer: FM: Oversampling in BETA3A, on 2nd CPU in v4.22. Todo: But what about non-FM?)
- Make CPU load configurable in Overlap=0 mode.
- Restore quality/latency display based on the new settings.
- Check the interface for settings that are no longer used (for example "Multiband HQ").
- Fix Stereo Boost meter (stops after few/many minutes/hours of work)
- Fix Stereo Boost meter (shows wrong levels, too low - sound is ok though) ■ OK
- Add GATING to compressors (stop increasing volume when the input volume gets below a threshold, reduces pumping and very loud volumes after a period of silence).
- Figure out what the difference is between AGC and what I'm doing now, see viewtopic.php?f=18&t=720&start=20 (among others). ■ DONE, but I need to add an extra slider to the user interface for 'remove remaining peaks'
- Improve Deep Bass Boost for higher frequencies (upto ~ > 100 Hz? Probably with a sliding scale?) See post Bojcha on page 40.

Later:
- Add "Phase Rotator" usefull for microphone processing
- Add Pre-Emphasize for Normal Output (10,15,20,25us), useful for web streaming (?)
- "Do you really want to close ..." message in stand alone version
- Add "LINK" checkbox for "phase differences" in "Calibrate FM transmitter..." (20Hz-10240Hz), [someone call it TILT]

Fantastic!! I'm Denny Stout ..I'm new here and I Just want to share this quote as my first post here...

"The less you can live on, the more chance your idea will succeed. This is true even after you’ve 'made it...

Make it a Great day people...

Cool... In my code I used things like

float a = .....;
float b = a * .5

Now looking at the assembly code that came out it turns out that the code that the compiler generates for this is:

        cvtps2pd  xmm5, xmm4                                    ;Convert FLOAT (32 bit) to DOUBLE (64 bit)
        mulsd     xmm5, xmm6                                    ;Multiply by DOUBLE value
        cvttsd2si esi, xmm5                                     ;Convert DOUBLE back to a single FLOAT
        ....

So that's 2 completely unnecessary (and expensive) conversions...

I'm now replacing all numbers like '.5' - which are apparently treated as doubles - by values like '0.5f' - which are treated as floats.

Right now I've converted Highpass filter, Final Limiter, Loudness and Hard Limit - and the CPU load of those 3 filters combined is now 6% lower! Unfortunately in the process I broke dualband, multiband, stereo boost and noise gate - so I'll need to fix those before I can upload a new version. But a CPU load reduction is coming soon - probably tomorrow.

Edit 2010/02/07: Dualband, Multiband and Noise Gate are now working again, but Stereo Boost isn't working properly for bass sounds... So no new version yet today

Edit #2: It seems that the bug was actually in the previous version... In the new one, Stereo Boost is much closer to the sound in v4.22 than in the previous BETA3(A)'s.
I still need to fix some minor problems and check where the difference in sound (at about -80 dB) between the last BETA3A and the new version comes from. CPU load seems to be about 4-5% lower than in the previous version, with everything enabled (FM Loud Bass Europe preset).

Updated BETA3A uploaded.

Changes:
- CPU load on average is about 4% lower than before. Audio effect of the rounding errors caused by this optimization is below -90 dB, with all filters enabled.
- Stereo Boost works much better now on low frequencies (much more similar to v4.22)
- Found and solved a bug in the previous BETA3A versions that caused clicking sounds at about -70 dB (they are gone now)
- Found and solved a rounding bug that caused noisy distortion at about -60 dB (probably not noticeable, but it's at least theoretically an improvement).
- Found and solved a bug in previous BETA3A's that caused the audio to be too soft at the start because the lowest bands of Multiband started at a very low volume.

NOTE: "Block Overlap > 0" does NOT work in this version!


Edit: Maximum processing speed is now 2% FASTER (better!) than in v4.22 for FM Loud Bass Europe (quality High in v4.22) preset. On top of that, in v4.22 part of the processing (FM oversampling) was done on a 2nd core, and in the new version everything is done on one core. Which means that the total CPU processing power that's used is considerably lower in this new version.

Hi Hans,
Great work..
Just to mention Highpass filter, it seems work more like notch filter . \ /

What do you mean by notch filter? And is that for the phase linear or non-phase linear processing?

- 4096, st_dsp, block overlap =0, phase-linear bandpass...
If i play 50Hz Sine and put Highpass to 50Hz it kills it. But if i continue to 150Hz for highpass it rise up 50Hz tone, so it's filtering only that freq.
30Hz sine tone, same thing, only difference tone is killed with 38Hz Highpass.
Non-phase filtering is not good.
beta3a from 01.feb and 29.jan doing same thing.

I just checked this not occure when block overlap is >0

Ah ok, that makes sense.

Non-phase linear filtering only works at 20, 30 and 40 Hz - I still have to add other frequencies.

Phase linear filtering is 'calibrated' to filter correctly at 30 Hz, and other frequency may give a different behavior.

For Overlap>0 I'm using the same filtering method that I used in v4.22, but with Overlap=0 that's not possible without getting quite bad artifacts. (At least I don't think it is possible, will take another attempt at it later).

Note: In the latest BETA3A, Overlap>0 does NOT work anymore for Noise Gate, Pre Limiter and Multiband - and there was already a problem with Hard Limit... (Highpass should still be fine though).