Tilt control

[Geoffrey Armstrong]

I only just got around to trying the Sweetone IR file you kindly provided @copy_of_a.

 

Within HQPlayer's convolution engine it nicely cures the "cold glassy" sound of Joni Mitchell's blue.

 

Do you also have a "light" version of this to cool down the sound of recordings that are too warm? … or could you easily generate one? That would be appreciated.

 

Using IR files within HQPlayer like this has the advantage of being available to DSD files as well, as @bogi mentioned.

 

Of course we might want to apply different settings via the VST plug-in depending on the recording. Some may require more "warming up" or "cooling down" than others. This can be done easily when your hosting the VST within a player app like J River or Audirvana; but there is the risk of latency echo (on windows in any case, from my experience). Otherwise though, we'd have to generate different IR files for different recordings, depending on how obsessive we are about applying the perfect setting for each recording.

 

It's interesting that Foobar on Windows doesn't have the latency "echo"  problem, according to @bogi.

 

I suppose there is always the risk of latency echo, whenever we're doing any DSP in real time. Although, I've never noticed it when using HQPlayer DSP. It might be there in the form of subtle added reverb, which I've either never picked up on, or didn't bother me.

 

The only way to be sure of avoiding that, I believe, is to do your DSP offline. This is why my use of MrsWatson is completely free of latency echo effects. I'm applying it to create temporary music files which have the tilt correction (using sTilt) applied. Those temp files are then sent onto HQPlayer and discarded after they've been played.

 

The IR files provided by @copy_of_a are already at 192khz/24bit, so there is no need to apply HQPlayer's expand HF option within the convolution engine.

 

As you mentioned @copy_of_a , "The correction curves are not "straight" but slightly shaped. Therefore it's not really a "Tilt" in a strict sense but a musical approach to "brighten" or "darken" music".

 

The sTilt developers acknowledge Quad on the page discussing their plug-in. You also pointed out that Sweetone's curves appear to correspond closely to those of Quad's filters. I don't know if sTilt is closer to a "straight line" Tilt control; but it doesn't have that added "sweetness" of Sweetone to my ears.

 

I haven't tried the slick EQ IR file you provided yet.

 

As @Miska points out in the HQPlayer manual; there are two convolution engines: "overlap-add" and "overlap-save". The second consumes more CPU power. He also warns about the demands of applying IR filters to multi-channel source material.

 

Obviously running the convolution engine is going to be placing greater demands on resources and, although I haven't encountered any problems, it might affect the ability to run the more demanding up-sampling filter and shaper combinations.

 

These problems can only be avoided by loading a pre-DSP music file to HQPlayer, so MrsWatson does have an advantage there.

 

Its most annoying limitation is that it's necessary to reduce the bit depth of any 24bit files to 16 bit before it can process them. It also needs a wav file for both its input and output. I achieve this by passing the original file through sox before sending the result to MrsWatson.

 

I emailed Teragon audio and the developer Nik Reiman came back to me to say; "sadly I do not have much time to work on MrsWatson these days".

 

The source code is available and under a BSD license, so I suppose another developer(s) could pick up on it and make the needed improvements.

 

 

[bogi]

9 hours ago, Geoffrey Armstrong said:

Its most annoying limitation is that it's necessary to reduce the bit depth of any 24bit files to 16 bit before it can process them.

 

Hi Geoff, are you sure about that? When I used 192/24 input file, I got 24bit file on output. OK, I didn't look if the samples are not reduced, but it was 24 bit WAV. The tool also has --bit-depth parameter and according  to command line help it can be even 32 bit.

 

C:\Users\bogi>mrswatson --help bit-depth
Run with '--help full' to see extended help for all options.
Quickstart for effects: C:\Users\bogi\mrswatson -p <plugin> -i <input file> -o <output>
Quickstart for instruments: C:\Users\bogi\mrswatson -p <name> -m <midi file> -o <output>

--bit-depth <argument> (default: 16)
  Bit depth to use for processing. If the input source specifies a bit depth,
  than that value will override the one set by this option. Valid values for
  bit depth include: 8, 16, 24, 32.

Please note the sentence: "If the input source specifies a bit depth,  than that value will override the one set by this option."

[Geoffrey Armstrong]

I will be very happy if you turn out to be right Bogi. Here is my email to the developer though, followed by his response:

 

“

I am trying to incorporate calls to MrsWatson in some apps I’m developing, running the sTilt 64 plug-in

 

On 16bit, 44.1 source WAV files this works fine. On 24bit, 96khz files though, I’m getting the following error:

 

UNSUPPORTED FEATURE: Compressed WAVE files

  This feature is not yet supported. Please help us out and submit a patch! :)

  Project website: https://github.com/teragonaudio/mrswatson

  Support email: [email protected]

 

Here is the command that resulted in this:

 

mrswatson64.exe --input "C:\jr-hqp\mrswatson\01 All I Want.wav" --output "C:\jr-hqp\mrswatson\01 All I Want-warm.wav" --plugin "sTilt x64,warm.fxp”

 

On another occasion I got the message that bit depths above 16bit are not supported.

 

It appears that any bit depths above 16 will not work and have to be reduced, e.g. from 24bit to 16bit for MrsWatson to work with them.”

 

 

 

 

Are there any plans to lift this limitation in the future?”

 

”

Hi Geoffrey,

Sadly I do not have much time to work on MrsWatson these days, so I don't know when these features will be implemented. I think your best bet is to use another batch program to convert the files to 16-bit WAVE.

 

Best,

Nik”

 

So it appears the developer is acknowledging this limitation. As you can also see, all I got was an error response when I tried to run the command on a 24/96 music file.

 

Have you tried it on actual 24/96khz or 24/192 khz files?

 

If so, and it worked for you, can you post here the mrswatson command you used to produce the output file?

 

It’s not clear to me from the commands you posted above, which music file you successfully processed.

 

Thanks

[bogi]

All the commands I posted a week ago processed the same 192/24 file and generated file out.wav.

Foobar2000 shows these properties of out.wav:

 

 

What surprises me is the message you got: "UNSUPPORTED FEATURE: Compressed WAVE files".

I didn't know that WAV files can be compressed, I never used such ones. But I searched web and found

WAV - Wikipedia

The WAV format supports compressed audio using, on Microsoft Windows, the Audio Compression Manager. Any ACM codec can be used to compress a WAV file.

 

I am curious how can one get such an exotic WAV? :o It seems to me that it could be enough to decompress your file (I don't know how) or create it once again if original source was something other (FLAC)?

[Geoffrey Armstrong]

Thanks Bogi. The original source was flac; but I converted it to wav using sox. So it should have created a regular uncompressed wav.

 

Can you tell me which version of MrsWatson you’re using? Just in case I’m on an older version.

[bogi]

MrsWatson version 0.9.8, build 20150122

 

[bogi]

I created my WAV with Foobar2000 Converter also from FLAC.

[Geoffrey Armstrong]

Ah, you've given me hope 😀 I just realized I downloaded from the Teragon Audio web-site the static version 0.9.7. I should have downloaded from github. I now also have the latest 0.9.8 version and will re-test with that.

 

Thanks

[Geoffrey Armstrong]

Well @bogi I'm very glad you turned out to be right 😀

My bad for using the older version from Teragon Audio, rather than the latest on gitHub.

Too bad the original developer, didn't put me straight on that.

 

Anyway, as you said, this handled 24bit files just fine using sTilt.

It's also quite clear how much I was missing by reducing the bit depth before sending the file to MrsWatson. Fortunately that's no longer necessary.

 

I did also try the new version with the Slick EQ and Sweetone plug-ins.

 

The only improvement I got there, was that with the SlickEQ plug-in, I'm no longer getting a file longer than the original (padded with silence).

 

The resulting file from MrsWatson is the same length as the original though.

 

Sadly though, I'm still getting the snap,crackle and pop with both those plug-ins.

 

So I'm sticking to sTilt, which works fine now.

 

I'll probably be using Sweetone using the IR through HQPlayer's convolution engine for DSD source files, only and for personal use. sTilt though MrsWatson is fine for me now for use with PCM originating files.

 

I'm not sure if any of my DSDs will need warming up though. More likely cooling down, in some cases.

 

Thanks again,

 

Geoff

[copy_of_a]

On 9/27/2020 at 11:28 AM, Geoffrey Armstrong said:

Do you also have a "light" version of this to cool down the sound of recordings that are too warm? … or could you easily generate one? That would be appreciated.

Here are 2 packs (192kHz/24bit wav):

 

- Slick EQ M Tilt in 0.3db steps (brighter + darker) from 0.3 - 1.5db

https://my.hidrive.com/share/julcgyvl0j

 

- Sweetone Tilt in 0.3db steps (brighter + darker) from 0.3 - 1.5db

https://my.hidrive.com/share/mgx0a6ucfs

 

 

On 9/27/2020 at 11:28 AM, Geoffrey Armstrong said:

This can be done easily when your hosting the VST within a player app like J River or Audirvana; but there is the risk of latency echo (on windows in any case, from my experience).

Again... there's something wrong with the conversion software. If VSTs (or other plugins) couldn't run with correctly compensated latency they couldn't be used in professional envirements for recording, mixing, mastering etc.

So what you are hearing is not an attribut of VST pugins but of the incorrect way their processing is handled in the respective host and/or processing software.

 

 

On 9/27/2020 at 11:28 AM, Geoffrey Armstrong said:

The sTilt developers acknowledge Quad on the page discussing their plug-in. You also pointed out that Sweetone's curves appear to correspond closely to those of Quad's filters. I don't know if sTilt is closer to a "straight line" Tilt control; but it doesn't have that added "sweetness" of Sweetone to my ears.

sTilt by default is fully "straight". Similar to Slick EQ's tilt in this regard.

The text of sTilts developers IMHO is just marketing - it has nothing to do with the dedicated design of the curves but more with the general purpose of Tilt EQs.

 

[Geoffrey Armstrong]

Thanks very much for those IR files. I’m looking forward to doing some critical listening with those.

 

I accept the latency issue is a problem of the apps hosting the VSTs and processing the files in real-time. They’re not actually DAWs of course. So not primarily intended for that purpose.

 

Still it’s curious that there seems to my ears to be much less of a problem (if any) on Mac.

 

Thanks for clarifying that sTilt and Slick EQ are straight line Tilt controls.

 

Geoff

 

 

[copy_of_a]

1 hour ago, Geoffrey Armstrong said:

Thanks for clarifying that sTilt and Slick EQ are straight line Tilt controls.

Here is a graph comparing Slick M (green) and Sweetone (white) at -0.6db tilt ("db" refers to db per decade, BTW).

Due to the shape of the curves (which are basically 2 shelf filters at the same center frequency - one with positiv one with negativ gain) Sweetone sounds a bit "warmer" while Slick M certainly sounds super "transparent".

The higher the correction the more the differences are emphasised, of course.

 

[Geoffrey Armstrong]

Thanks for sharing that.

 

I would also love to learn about recordings you or anyone else here has come across that are either too cool/bright, warm/dark that should respond well to Tilt correction.

 

I will try to add some more of my experiences over the next few days.

 

If we can share examples it could save us all some time.

[bogi]

I attempted to create IRs from Sweetone and compared it with the result of @copy_of_a.

I alternatively tried MRsWatson and Foobar2000 to get sine sweep response and then for both cases I used Voxengo Deconvolver to deconvolve the result of previous step. For both I used the same sweep sine generated in Voxengo Deconvolver. It was 60s 192k 24bit sine sweep with fade in / fade out option for Voxengo's "reverse" technique.

 

Method 1) with MrsWatson

1. In Audacity I loaded the plugin, disabled graphical interface and set desired values, then saved .fxp preset

 

2. Called MrsWatson and applied .fxp preset to the faded sweep, for example:
C:\Users\bogi>mrswatson --bit-depth 24 --plugin Sweetone,Sweetone-0.6.fxp --input H:\Documents\testtones\Sweep-192000-24-M-60.0s-fade.wav --output Sweep-192000-24-M-60.0s_resp-fade.wav

 

3. Loaded both the faded sine and the response in Voxengo, see picture

 

It generated file Sweep-192000-24-M-60.0s_resp-fade_dc.wav, which can be then renamed as you wish and loaded into HQPlayer Convolver.

Method 2) with Foobar2000

1. I dropped the test sine into Foobar2000 playlist

2. Right clicked it and chose the Converter ... in context menu.

3. In the Converter window I set output format WAV, Dither (always), chose the destination folder and clicked on Processing

4. In the Processing dialog I moved the plugin to Active DSP and clicked Configure Selected.

5. I used the plugin graphical interface to set the tone tilt value as you see on the picture (and turned off preamp on the right side):

 

6. I closed the plugin GUI, clicked Back in the Processing dialog and pressed Convert button in Converter dialog.

 

7. I loaded the resulting WAV response into Voxengo Deconvolver and processed it in the same way as in the point 3 of method 1).

 

I know that the tilt levels are for sure not the same, since in graphical GUI I could not set numerical value. But when I compared my results of 1) and 2) I concentrated more on the resulting sound quality. I found the Foobar way better. Here are my results: IR.zip
I found the IRs from copy_of_a best in regard of sound quality but the Foobar2000 result is IMO good and usable if I will want to process some other plugin.
The IRs from copy_of_a load longer time into HQPlayer so it seems they are of better resolution. Maybe I could use longer than 60 sweep or maybe switching to the latest JRiver MC, which supports 64bit VST3 plugins, could yet improve the result quality. What do you think about my procedure and results?
 

[bogi]

@copy_of_a, is it possible to generate such a graph like you posted above from any IR file? Is it possible to look at my result, if it is something crazy or something quite usable from technical point of view?

 

[copy_of_a]

4 hours ago, bogi said:

@copy_of_a, is it possible to generate such a graph like you posted above from any IR file? Is it possible to look at my result, if it is something crazy or something quite usable from technical point of view?

I have to admit I don't understand all details of your process but I've looked at your files and made a graph for you in REW.

Unfortunately your processed files are not usable.

The audio files are way too short and I assume actual audio information got lost.

You should leave some silence before and after the actual dirac impulse.

I use 1.5 seconds before and after and HQPlayers convolution engine can read those IRs without issues.

You also seem to have applied a wrong setting with Sweetone... see below your MrsWatson Sweetone -1db (red) and my Sweetone -1.5db (green).

I'm not familar with REW (only use it for analysing some stuff from time to time) but I think you can create IR files with Tilt correction applied through its internal equalizer. REW is also free.

As far as the parameters in Sweetone go you can display and edit the values when you right-click on the gain knobs... see below.

Hope that helps.

 

 

[copy_of_a]

double post

[bogi]

Thanks! I will try your suggestions.

[Geoffrey Armstrong]

Once again @copy_of_a I really appreciate the trouble you went to creating those IR’s with different values.

 

I realized that by using HQPlayer’s convolution engine with these, not only can you use them for DSD; but also for anything you can play through HQP, including Qobuz and Tidal streams.

 

That makes it easier for me to identify, which albums I’ve been streaming lately, that I feel could use some Tilt correction.

 

From listening I believe the sTilt corrections I’ve used so far, correspond roughly to your .012 db or even .015 db corrections, in either direction using skick EQ and Sweetone.

 

I feel I may have gone a little OTT at times; but I’m enjoying the results anyway.

Duke Ellington’s Jazz Party, has some wonderful percussion, as mentioned on Stereophile, which really stands out.

http://open.qobuz.com/album/0074644071221

 

I prefer it warmed up at .012db, even though it feels a bit of a guilty pleasure, like eating too much chocolate. Those percussion instruments appear to stand out just as clearly, and now there seems to be more athmosphere around them.

 

Horns may seem slightly muted now; but I’m loving the rich burnished textures.

 

The same remarks apply to Hero Trio, also discovered via Stereophile recently. http://open.qobuz.com/album/lg6mqlw1f0sub

 

For both of these I feel I really ought to back off a little; but I’m enjoying them too much warmed up as much as I have.

By contrast the latest one from Diana Krall definitely sounds to me like it needs cooling down. I’m enjoying the greater sense of transparency and detail Cool Tilt brings to this one and the same goes for most of her œuvre.

 

This reminds me that it’s often an aesthetic choice by those responsible for a recording. Just like a models skin may be warmed up by a photographer, with all the tools at their disposal.

 

Incidentally, I accidently played part of the first track of her new album warmed up! It was unlistenable. It shows that it’s quite obvious when you’ve warmed up an album that is already too warm or cooled down one that’s already too cool.

 

The aesthetic choice to make an album sound ultra transparent, regardless of what it’s played back on is probably the more common situation.

 

So there are probably more recordings that benefit from warming up. Even if you go a little too far in that direction, the results are usually pleasing, like listening through certain tube amps.

 

Initially I thought this one needed warming up a little:

 

http://open.qobuz.com/album/0614427973023

 

Natalia Mateo’s voice is so transparent, you might think it’s slightly too cool. I came to the conclusion that this is natural transparency though, rather than a faux transparency, that is all too common.

 

So this is a good example for me of learning to leave well enough alone.

 

With a new toy there’s obviously a temptation to play with it all the time.

 

I’m gaining confidence in determing what treatment, if any, each album needs.

 

It soon becomes obvious.

 

Tilt correction is a very effective form of correction when called for.

 

 

 

[bogi]

Some of my recent results regarding VST plugin to IR conversion.
I used 6s 192/24 linear sweep generated by Voxengo Deconvolver with or without silence on one or both ends. I used the sweep with added silence only when "playing it" through VST plugin to get WAV response. But after that in Voxengo Deconvolver I used the sweep test tone without the added silence as a reference to compare the WAV response with. I think that corresponds to more usual scenario when sweep is played through loudspeaker or music instrument and response is recorded using microphone. You stop recording a second or two later as the test sweep playback finished.

Voxengo Deconvolver was always used with checkbox [x] Normalise to -0.3 dBFS
Slick EQ -1.5dB setting for the following:
1) curved lighter green: full volume sweep without silence on begin or end

2) blue: half volume sweep with 1.5s silence at both ends

3) dark red: half volume sweep with 1.5s silence at end, but in this case I used sweep including silence at end as reference in Voxengo Deconvolver (otherwise I used sweep without added silence as reference when deconvolving)

4) the darker green (near the same as the dark blue one): half volume sweep without silence on begin or end

Result: full volume of input sweep leads to unusable results. Silence at beginning destroys the result. I tried it more times but it always caused disaster with Voxengo.

 

When 1.5s added silence at end of sweep (no silence at beginning) was used to generate WAV response, but then sweep without added silence was used as reference during deconvolving:

Slick EQ -1.5dB and -1.0 dB setting are darker green and blue one.

The curver light red and light green are Sweeone -1.5 and -1.0.
Note the vibration at beginning - it occurs only if the recorded WAV response contains also 1.5s of silence response at end, but the test signal used in Voxengo Deconvolver was pure sweep without added silence at end.

When I used added silence also for test sweep in Voxengo Deconvolver, then those vibrations did not occur - see the dark red one on the previous graph - some very very small vibration around the dark green one can be seen in middle region on the 1st graph.

 

Now the same half volume of input sweep as in the previous picture, but without the 1.5s silence at end of test tone. No silence was used at any part of the process:

So the silence at end, when it was used only to 'record' the WAV response but not in reference test tone during deconvolving, is causing the vibrations on the previous example, but on other side it is keeping the frequency response at higher levels in bass region.

I didn't try different lengths of silence at end. I tried also -12 dB / octave high cut after 24kHz in Voxengo, but it didn't have visible effect in teh 20Hz - 20kHz region - the vibrations at beginning were still here.

 

@copy_of_a please what do you think about it?

[bogi]

I was able to get the expected frequency response when I did the following:

I used two sweep test tones. The one generated by Voxengo I used as test tone in Voxengo Deconvolver. But for creating response WAV (by applying VST plugin to the test sweep) I needed to use a bit modified test sweep:

• I lowered volume of test sweep to about 80% or less (90% already brought oscillations like the blue one on the 1st picture of my previous post).
• I had to insert 3 seconds silence after the volume lowered test tone (1.5 seconds were not enough). That helped to get rid of the small oscillations in the bass region like in the middle picture of the previous post. Any leading silence destroyed the result so I used no leading silence.

The result for SlickEQ -1dB I got this way is the blue one. It is pretty linear from 30Hz (near the maximum dB level) to much more than 20kHz.

The orange one is the same but with only 1.5s silence after the test tone (with other words, recording of test tone response was stopped 1.5s after test tone finished in comparison with 3s delay used in the blue case).

The green one differs from the blue one in using test tone with 3 seconds silence as test tone in Voxengo Deconvolver during deconvolution. In the blue case I used the reference test tone without added silence.

 

 

The blue case result plays well in HQPlayer, here it is: SlickEQ-1.0dB.wav.zip
The Voxengo Deconvolver settings used are:

 

 

Next I can experiment to get best possible sound quality, for example:

- by using longer test tone than 6 seconds

- by comparing different VST host solutions (MrsWatson vs. Foobar2000 vs. JRiver MC vs ...)

Any suggestions are welcome.

[Miska]

Remember that you can create various sets of different corrections (and plain no-correction) as Matrix profiles in HQPlayer. So instead of using the simple convolution engine, you can setup the same in pipeline matrix and then create profiles out of those. Then it becomes pretty easy and quick to switch between different corrections.

 

[Geoffrey Armstrong]

5 hours ago, Miska said:

Remember that you can create various sets of different corrections (and plain no-correction) as Matrix profiles in HQPlayer. So instead of using the simple convolution engine, you can setup the same in pipeline matrix and then create profiles out of those. Then it becomes pretty easy and quick to switch between different corrections.

 

That’s a great tip, thanks @Miska I noticed there’s a Matrix option in HQPD on IOS; but couldn’t understand what it does. Perhaps by setting up profiles as you’ve described, we can switch between them with HQPD?

[bogi]

Thanks, Miska. I don't know if convolution setup is saved when you export settings from HQPlayer Desktop. But when the wish is only to set up different IRs for convolution, it is easier to switch Matrix profiles only. Matrix profiles can be switched also from command line with hqp-control2:
--matrix-list|--get-matrix|--set-matrix <name>

[bogi]

4 hours ago, Geoffrey Armstrong said:

That’s a great tip, thanks @Miska I noticed there’s a Matrix option in HQPD on IOS; but couldn’t understand what it does. Perhaps by setting up profiles as you’ve described, we can switch between them with HQPD?

 

Matrix pipelines can be used with or without IR WAVs.

An example without IRs is multichannel to stereo processing for listening through loudspeakers - from HQPlayer manual:

Another more complicated example is to use HRTF (Head Related Transfer Function) IRs to convert stereo or multichannel audio to binaural sound for listening on headphones. You can find such examples here:

https://audiophilestyle.com/forums/topic/19715-hq-player/page/283/#comments