Ending the Windows Audio Quality Debate

[Offler]

Just see daftcombo's iZotope analyses of the example files I posted. They look like the same in FFT view, but the waveform peak values are vastly different, and therefore the methodology of look at individual FFT bins for peak level alignment is flawed:
https://www.audiosciencereview.com/...indows-audio-quality-debate.19438/post-890577

Nope, we are still not talking about the same thing.

I was trying some tests using dual sinewave to find out about IMD of my amplifier, instead I encountered distortion up to -60dBFS, which later proven to be audible.

Previously i confirmed that no matter what you do in EQ APO, if you set sinewave volume to -0,12dB or more, CAudio Limiter will trigger and no higher volume setting is possible. Instead the amplitude will be normalized and thus distorted, compared to the original.

This is visible on FFT graph as a peak which will not rise when volume is increased in EQ APO, only if you increase the signal on the amplifier. Its also visible on oscilloscope as the amplitude will not rise beyond certain threshold.

The difference is, that frequency and phase for two signals is in my case fixed by the software i use. Thats why i said i dont have tools to test more than two signals at the time. I am not using a file, but a tone generator.

So to filter out the distortion caused by CAudioLimiter I had simply to understand that I am mixing two signals, and therefore the threshold for triggering Limiter is 6dBFS lower than I originally expected it to be. Once I sorted that out, distortion was gone - I just needed to understand why it was triggered once again.

So sorry to say, but your posts are in no relation to what I was doing and reading from FFT graph.

 

[bennetng]

Nope, we are still not talking about the same thing.

I was trying some tests using dual sinewave to find out about IMD of my amplifier, instead I encountered distortion up to -60dBFS, which later proven to be audible.

Previously i confirmed that no matter what you do in EQ APO, if you set sinewave volume to -0,12dB or more, CAudio Limiter will trigger and no higher volume setting is possible. Instead the amplitude will be normalized and thus distorted, compared to the original.

This is visible on FFT graph as a peak which will not rise when volume is increased in EQ APO, only if you increase the signal on the amplifier. Its also visible on oscilloscope as the amplitude will not rise beyond certain threshold.

The difference is, that frequency and phase for two signals is in my case fixed by the software i use. Thats why i said i dont have tools to test more than two signals at the time. I am not using a file, but a tone generator.

So to filter out the distortion caused by CAudioLimiter I had simply to understand that I am mixing two signals, and therefore the threshold for triggering Limiter is 6dBFS lower than I originally expected it to be. Once I sorted that out, distortion was gone - I just needed to understand why it was triggered once again.

In your case, better use something capable of generating audio files so that you can inspect and analyze the original waveform offline, in order to make sure the waveform peaks are low enough to avoid CAudioLimiter. My test files for example are generated by SoX, and it can generate a lot of tones, with adjustable phase.

So sorry to say, but your posts are in no relation to what I was doing and reading from FFT graph.

100% related. Because you need to understand that FFT plots cannot reliably reflect peak level, and knowing the peak level is important if you want to avoid CAudioLimiter.

 

[daftcombo]

In your opinion, how many more dBFS can produce a high-pass filter on music? Let's say a 96dB/octave slope put at 25Hz. In the case of a minimum phase filter? in the case of a linear phase (FIR) filter?

I use such a FIR filter to relieve my Aria 906 of frequencies they wouldn't be able to play anyway, applying a -4dB preamp in EqAPO as suggested by OP. So far, so good.

 

[bennetng]

In your opinion, how many more dBFS can produce a high-pass filter on music? Let's say a 96dB/octave slope put at 25Hz. In the case of a minimum phase filter? in the case of a linear phase (FIR) filter?

I use such a FIR filter to relieve my Aria 906 of frequencies they wouldn't be able to play anyway, applying a -4dB preamp in EqAPO as suggested by OP. So far, so good.

To be honest, I don't know. However, if you can export the filter impulse you prefer to use, you can import the impulse into for example, foobar's convolver, and then use foobar's file converter to render your audio files offline, then you can inspect the waveform peaks with your iZotope tools for instance. If want to do this on a lot of files and worry about wearing your SSD, you can render to text files with BitSort:
https://www.audiosciencereview.com/...he-obsession-with-dr-meters.11297/post-649576
The readme file explained how to use BitSort with foobar2000. Add the Convolver into foobar's file converter DSP chain, choose 32-bit output, and BitSort will show the audio files' peak level, including 0dBFS+ values.

 

[daftcombo]

To be honest, I don't know. However, if you can export the filter impulse you prefer to use, you can import the impulse into for example, foobar's convolver, and then use foobar's file converter to render your audio files offline, then you can inspect the waveform peaks with your iZotope tools for instance.

I tried that already. In Foobar2000, I took a few tracks, used "Convert" with a Convolver step with the impulse WAV I use in the chain, and saw what it gives in iZotope. The sound wave profile was a bit lower in amplitude (there's already a little pre-amp in the impulse file) and the dB peak was lower. But I couldn't deduce the value from the original peak value and the pre-amp, and wanted to know if there's a formula.

 

[bennetng]

I tried that already. In Foobar2000, I took a few tracks, used "Convert" with a Convolver step with the impulse WAV I use in the chain, and saw what it gives in iZotope. The sound wave profile was a bit lower in amplitude (there's already a little pre-amp in the impulse file) and the dB peak was lower. But I couldn't deduce the value from the original peak value and the pre-amp, and wanted to know if there's a formula.

I can't answer this question, but I missed something in my previous reply. Apart from the Convolver, also make sure the output sample rate is identical to your real time playback environment. For example, if you use 96kHz, add a resampler in the DSP chain as well, and don't use SSRC as it is ridiculously steep (similar to Chord), which may generate excessively high peaks and therefore not a good candidate to approximate the Windows resampler. The RetroArch resampler with highest quality settings, or the SoX resampler plugin with default settings make more sense.

 

[daftcombo]

I can't answer this question, but I missed something in my previous reply. Apart from the Convolver, also make sure the output sample rate is identical to your real time playback environment. For example, if you use 96kHz, add a resampler in the DSP chain as well, and don't use SSRC as it is ridiculously steep (similar to Chord), which may generate excessively high peaks and therefore not a good candidate to approximate the Windows resampler. The RetroArch resampler with highest quality settings, or the SoX resampler plugin with default settings make more sense.

That's what I always do. You are right that it is compulsory to put a resampler in Foobar2000 before Convolver in the chain.
I use SoX, target sample rate=44,100, quality=Best, passband=95%, phase response=50% (linear).

 

[kernelpanic]

So regarding the -6.30db setting, would that be pre or post EQ? For example, I have EQ APO preamp set to -6.30db, but when you factor in the EQ I have applied, the resulting gain is -4.2db.

 

[daftcombo]

So regarding the -6.30db setting, would that be pre or post EQ? For example, I have EQ APO preamp set to -6.30db, but when you factor in the EQ I have applied, the resulting gain is -4.2db.

View attachment 149763

View attachment 149766

It should be post EQ. You still need to substract 2.1dB.

 

[Offler]

100% related. Because you need to understand that FFT plots cannot reliably reflect peak level, and knowing the peak level is important if you want to avoid CAudioLimiter.

But i wasnt using levels in FFT to read real peak levels. I was using it in laboratory-like test to see whether the level of artifically generated sinewave rises or not beyond certain gain values in EQ APO.

Its just a way to confirm that the distortion is for sure caused by CAudio limiter.

So regarding the -6.30db setting, would that be pre or post EQ? For example, I have EQ APO preamp set to -6.30db, but when you factor in the EQ I have applied, the resulting gain is -4.2db.

View attachment 149763

View attachment 149766

Well, it depends on few factors.

1. If you play music on PC using Wasapi Exclusive mode or ASIO, you would not encounter this type of distortion.(But you lose equalizer as well).

2. Lets assume you want to improve on audio quality of games and browser.
Then assume Windows is typically mixing up to 2 sources. If you like to play music from internet browser, or you like to play games, and use soundtrack of your own - 'Peak gain' in 'Estimated properties' should be -6.15dB or less, ideally -6.3.

Also the recommendation was already mentioned - turn off all windows sound, and if possible mute them in Mixer as well, else you will get to 3+ sources sooner than expected.

Last assumption is that games and browsers use to mix sounds on their own, as most games use multiple samples at once and the internal mixer of the game isnt botched...

Good news is that most sound engineers are not maxing out the peak levels of the samples used. Sorta good news is that much of music /sound in streaming services is already normalized.

If you are using just one audio source at a time, peak gain at -0.14dB should be ok.

 

[Robbo99999]

@amirm Could you please consider making this thread pinned?


@DDF Just to make sure I followed up correctly, is this is what EAPO's configuration should look like? (With an EQ configuration file).

Just a -4 dB pre-amp to avoid up-sampling artifacts.

View attachment 106840

How late am I to this thread! You can also achieve the same thing by running a permanent "less than 100 percent" on the main Windows Volume Slider. So -4dB is equivalent to 76% on the Windows Volume Slider:

I personally assure -2dBFS for intersample overs and an additional -2dBFs for my temperamental Soundblaster G6 DAC that clips in the bass if you go over -2dBFS (as proven by Amir's review), so I allow a total of -4dBFS but for different reasons to you.

EDIT: I've previously measured my loudest tracks using Orban Loudness Meter and generally the highest oversample overs were +2dBFS, so that's why I allowed -2dBFS for the intersample over variable.

 

[daftcombo]

How late am I to this thread! You can also achieve the same thing by running a permanent "less than 100 percent" on the main Windows Volume Slider. So -4dB is equivalent to 76% on the Windows Volume Slider:
View attachment 152954
I personally assure -2dBFS for intersample overs and an additional -2dBFs for my temperamental Soundblaster G6 DAC that clips in the bass if you go over -2dBFS, so I allow a total of -4dBFS but for different reasons to you.

Perhaps you wouldn't need the extra -2dBFS if you used EqualizerAPO to disable "hidden APO's" as the OP suggested. It's worth a try probably.

 

[Robbo99999]

Perhaps you wouldn't need the extra -2dBFS if you used EqualizerAPO to disable "hidden APO's" as the OP suggested. It's worth a try probably.

What you mean, shut off "original APOs" as the orginal poster said? I don't know much about that effect. Are you saying that if I reran my loudest tracks on Orban Loudness Meter whilst those "original APOs" were shut off then they wouldn't trip into the +2dBFS intersample overs? To be honest I don't need that extra headroom as I've got plenty of volume headroom available, plus I'm not sure what other effects turning that option on has.

 

[Monstieur]

Has anyone confirmed whether post-mix in EAPO is sufficient to avoid CAudioLimiter? I'm current running -0.2 dB in pre-mix and would like to use post-mix instead.

If the application outputs 32-bit float and CAudioLimiter is applied only at the end, post-mix should be enough. However if an application like iTunes outputs 24-bit integer, I wonder if it CAudioLimiter kicks in during pre-mix. The playback device is only 24-bit integer.

 

[Monstieur]

A corollary questions is: If the playback device is 32-bit float, and the WASAPI chain is also 32-bit float, is CAudioLimiter bypassed completely?

 

[Offler]

Has anyone confirmed whether post-mix in EAPO is sufficient to avoid CAudioLimiter? I'm current running -0.2 dB in pre-mix and would like to use post-mix instead.

If the application outputs 32-bit float and CAudioLimiter is applied only at the end, post-mix should be enough. However if an application like iTunes outputs 24-bit integer, I wonder if it CAudioLimiter kicks in during pre-mix. The playback device is only 24-bit integer.

I have seen 32bit Integer output devices, but not FLOAT. Been looking for some just out of curiosity, but no luck.

Fix has to be applied in pre-mix because the distortion appear to happen during mixing process. That is also why it does not matter what parameters the output device has.

 

[daftcombo]

What you mean, shut off "original APOs" as the orginal poster said? I don't know much about that effect. Are you saying that if I reran my loudest tracks on Orban Loudness Meter whilst those "original APOs" were shut off then they wouldn't trip into the +2dBFS intersample overs? To be honest I don't need that extra headroom as I've got plenty of volume headroom available, plus I'm not sure what other effects turning that option on has.

You wrote that your soundcard clips the bass sometimes. Isn't that the effect of a hidden APO?

 

[phoenixsong]

Beginner to software/API stuff here! Is WASAPI native to all windows pcs? If not, what is being used on them?

I chanced upon this thread and it really piqued my curiosity, for I have always focused on hardware and merely adjusted windows audio settings to what extent I knew (No sounds scheme selected, system sounds and startup sound muted, spatial sound and all other enhancements disabled, 24 bit bit depth and music's original sample rate set, Windows Media Player 24 bit playback enabled and equaliser disabled).

I mainly listen to CDs and WAV files and youtube from my Windows 10 Pro laptop through my Motu M2 via USB connection, without any EQ engaged. Will Equaliser APO still be beneficial to me as the first post described? If it will, will it mess up the Motu interface's functioning with recording software? I definitely will read through this entire thread after my next few waves of assignments are over, but my acute hunger to know asap will keep me up at night unless I ask for some quick answers first

 

[Robbo99999]

You wrote that your soundcard clips the bass sometimes. Isn't that the effect of a hidden APO?

The clipping in the bass, which I haven't tried to experience, is outlined in Amir's review of the Soundblaster G6:
https://www.audiosciencereview.com/...w-and-measurements-of-sound-blasterx-g6.7016/

I think it's a hardware issue rather than software, I just make sure I'm running at least -2dBFS to account for the G6 DAC clipping issue, then an extra -2dB to account for intersample overs because I don't want the intersample overs to enter the clipping zone of the G6 - might not be necessary for me to go to this extent, but I've got the volume headroom and I'm just taking the approach of optimising everything to the best theoretical extent.

 

[Propheticus]

Beginner to software/API stuff here! Is WASAPI native to all windows pcs? If not, what is being used on them?

I chanced upon this thread and it really piqued my curiosity, for I have always focused on hardware and merely adjusted windows audio settings to what extent I knew (No sounds scheme selected, system sounds and startup sound muted, spatial sound and all other enhancements disabled, 24 bit bit depth and music's original sample rate set, Windows Media Player 24 bit playback enabled and equaliser disabled).

I mainly listen to CDs and WAV files and youtube from my Windows 10 Pro laptop through my Motu M2 via USB connection, without any EQ engaged. Will Equaliser APO still be beneficial to me as the first post described? If it will, will it mess up the Motu interface's functioning with recording software? I definitely will read through this entire thread after my next few waves of assignments are over, but my acute hunger to know asap will keep me up at night unless I ask for some quick answers first

Yes, it's part of the Windows audio architecture. It's actually a recommended low-level API for applications to stream audio with. DirectSound is deprecated.