Stereo Tool

This is definitely possible if the Trumpet filter kicks in too soon, and I would expect it to do something like this to female voices... I'll post a version WITHOUT Trumpet filter, to see if that's better.

Basically, what the Trumpet filter does is:
- It measures if there are highs with repeating spikes. For example, every 50 samples or so a loud spike occurs, which happens in trumpet sounds, but also in female voices.
- Then it measures the bass level.

The idea is this: If there's a lot of bass, it will cut off those spikes, which may cause some distortion (highs are gone for short periods during bass peaks).

The trumpet filter REPLACES this distortion by different type of distortion (!), which is probably what you hear:
- The spikes are cut off, and the overall volume is increased a bit to make up for that. This increases 'female voices' distortion that you hear with high Loudness values.


Now the fix I made should not change the Trumpet filter but... Previously due to the errors in the filter, I don't know what it did. It may have mis-behaved, which - apparently - had a positive effect on the sound. I'll post a version without Trumpet filter in about an hour, then I'll wait for your response. If the sound is better, I might need to reconsider even having the Trumpet filter - or at the very least set it far less aggressively.

After doing some more tests I've now come to the conclusion that Overlap=0 sounds better than Overlap>0 after all. (I've compared bass sounds, and the difference there is really huge). In Overlap=0 mode the bass level is a bit higher though, which can hurt the sound quality - but that can be fixed by lowering the bass level a bit, which makes the difference with Overlap>0 even bigger. And that's only the case in lower latency settings - at 4096 the difference is extremely small.

I see what you mean. I suppose the channel separation is not good enough. Will look into it.

ok, but i must listen day or two ...

in overlap=0 bass is stronger and qualiti better, but i noticed that sometimes in rare cases knows to hurt mid-lows...
Actually when i listen radio sometimes vocals from song around 150-500Hz "vibrates" beacuse of bass.
But again, bass can be set lower a bit by multiband or eq.

The combination of louder, less deformed bass and hurting mid-lows makes sense. Also, the CPU load in Overlap=0 mode is half that of Overlap>0 mode (with maximum overlap size). So I'm going to stop supporting Overlap>0 mode - I think in general it sounds better (given your comments I suppose you agree?)

By the way, I've heard what you're describing (the "vibrating" of 150-500 Hz vocals) in one track, but I also hear it (although slightly less) in Overlap>0 mode. Hint: Lowering Clean/Punch helps here! (I suppose I need to set the default value a bit lower, will check that in a while).

The versions WITHOUT Trumpet filter (For other readers: please use this for testing only - it might be sub-optimal but I want to know if it solves the problem):
Winamp DSP plugin: http://www.stereotool.com/download/dsp_ ... _BETA2.exe
Stand alone version: http://www.stereotool.com/download/ster ... _BETA2.exe
VST version: http://www.stereotool.com/download/vst_ ... _BETA2.dll


Edit #2: For female voices with 'vibration': Setting the highpass filter to 'non phase linear' seems to help also! Probably because it filters low frequencies out much stronger.

Edit #3: I've just done some comparisons of Overlap=MAX vs Overlap=0, and I didn't hear any difference in these female voice vibration cases. Are you sure that it's less for Overlap>0? (I've tested at Latency 4096, FM Loud Bass Europe preset, with Loudness set at 4.00, latest BETA3A version.)

Edit #4: The Trumpet filter was made to remove exactly this type of distortion, but it doesn't work at such low frequencies. So in this version without trumpet filter, you might hear more of this (and less of other types of distortion).

I'm running mediamonkey to play the music to Virtual Audio Cable, and then Stereo Tool stand-alone to process the sound to my sound card. However, I do not output the fm signal, just the normal signal. What you were suggesting was right, it is the pre-emphasis causing the highs in the 512/1024/2048 settings. If I turn it off, the high sounds are gone in these modes. But... Why is the pre-emphasis running if I output 'normal output' and disabled 'fm stereo/rds sound card' output?

So in other words, fm output is bypassed in 4096 when on normal output, but not in 512/1024/2048 it seems?
*EDIT* 4096 is partly bypassed, still sounds different with fm on / off

Because you might send the 'normal output' into an FM transmitter which performs pre-emphasis again. The processing has in that case been done on a pre-emphasized signal, so you won't get too high peaks in the output. Or, one might want to test the sound quality before turning the FM transmitter output on. If you don't want pre (and de) emphasis to occur, turn it off... That will give you a lot more highs btw.

Question: Why did you turn pre-emphasis and FM processing on in the first place? (Is something unclear in the interface, or were you using the FM presets, or ...???)
No, but de-emphasis doesn't work correctly (not strong enough) in lower latency settings.

Aha I thought normal output was bypassing the fm signal (because I'm hearing stereo, while the fm processor is on), and the fm output mode was normal output that was processed by the fm signal processor. I often switch between laptop speakers, and stereo (through fm transmitter), so that's why I leave it on, and only uncheck/check the checkboxes in the sound card configuration...

So in other words 'normal output' suggested that it is not fm output, if there is also fm output.

Like this:


(my ms paint skills are not the best lol)

And I understand what you're saying, but I think 'normal output' should always be aimed for direct speaker output, and fm output should be for fm transmitters. With an option to let the processing be done by the transmitter or by stereo tool (so basically an option for the fm output to send the de-emphased stereo signal, or the processed fm signal)

IF FM output is on, that would mean that I would have to process everything twice (once for FM output and once for non-FM output). Which would - among others - multiply the CPU usage by 2...

But there's an item on my TODO list requested by Bojcha (lower pre-emphasis values for normal output, to protect encoded MP3s against overshoots). When I add that I can probably immediately include something like "bypass FM pre-emphasis (if FM output is disabled)" or something like that.


I like your drawing skills in your avatar more but the drawing is understandable so...

Is the pre-emphasis filter not the final step?
Sorry, didn't have any decent drawing software :p