ReClock and Audio Discussion

[dansrfe]

For people using Reclock: What is the optimum sample rate to set in Reclock's configuration settings? The Realtek chip on my motherboard supports 24-bit, 192Khz so I have 24-bit padded to 32-bit and 192Khz set in Reclock but I'm not sure if that's how it should be.

Also, has anyone experienced problems with the audio "cutting" when using PAL slowdown with time stretch and 192Khz? I have to set the output sample rate in Reclock to 96Khz in order to get a consistent audio stream...

[e-t172]

For people using Reclock: What is the optimum sample rate to set in Reclock's configuration settings? The Realtek chip on my motherboard supports 24-bit, 192Khz so I have 24-bit padded to 32-bit and 192Khz set in Reclock but I'm not sure if that's how it should be.

Use 48 kHz. Higher sample rates are snake oil and do not present any audible benefit. (http://people.xiph.org/~xiphmont/demo/neil-young.html)

Also, has anyone experienced problems with the audio "cutting" when using PAL slowdown with time stretch and 192Khz? I have to set the output sample rate in Reclock to 96Khz in order to get a consistent audio stream...

From experience I recommend disabling time stretching as it often causes audible artifacts.

[secvensor]

How to disable reclock?

[iSunrise]

For people using Reclock: What is the optimum sample rate to set in Reclock's configuration settings? The Realtek chip on my motherboard supports 24-bit, 192Khz so I have 24-bit padded to 32-bit and 192Khz set in Reclock but I'm not sure if that's how it should be.
Like e-t172 already said, use the same bit accuracy and sampling rate (bandwidth) that your source has. Don´t go any higher, don´t go any lower. If what you´re doing is only playing back, you don´t need any headroom, because you´re not going to process it further. That is assuming that you´re using WASAPI, so there´s also no additional mixing involved. Purely for playback, I prefer 48KHz for almost everything (44.1KHz when listening to CDs). For movies and since you´re dealing mostly with encoded data, use 24bit to be safe.

[leeperry]

Like e-t172 already said, use the same bit accuracy and sampling rate (bandwidth) that your source has. Don´t go any higher, don´t go any lower.
Well, resampling to the same sample rate as the source will butcher the trebles IME/IMHO. It's measurable too: http://forum.slysoft.com/showpost.php?p=227357&postcount=3996

The sweet spot is prolly 88.2kHz if your audio device supports it. This said, those measurements were made with the old Reclock resampler and that will depend on the oversampling rate of your DAC chip as well(only the AKM chips such as AK4396 can go 128X all the way up to 192kHz, most other chips will fall back to 64X at 96kHz and 32X at 192kHz).

It's all discussed there really: http://www.audioholics.com/education/audio-formats-technology/upsampling-vs-oversampling-for-digital-audio

Ideally you wanna keep the oversampling rate as high possible and yet upsample one "notch" higher than the input sample rate so 88.2kHz is prolly the sweet spot. 96kHz will do if your gear doesn't support it.

Dan Lavry confirms it too: The Optimal Sample Rate for Quality Audio (From: http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf)
88.2 KHz and 96 KHz are closest to the optimal sample rate.
Also, jitter is usually lower at 96kHz than at 48khz: http://www.cirrus.com/en/pubs/appNote/AN339REV1.pdf
Sample Rate with CS8416 in low jitter mode:
48kHz: 122.57ps
96kHz: 45.278ps
And it might be worse for 44.1kHz multiples because DSP/receivers masterclocks are usually 48kHz multiples(24.576Mhz=48*512 for instance).

[iSunrise]

Well, resampling to the same sample rate as the source will butcher the trebles IME/IMHO. It's measurable too: http://forum.slysoft.com/showpost.php?p=227357&postcount=3996...
Yes, you´re right, not going to argue against it. I totally forgot that reclock also resamples, because it has to match the audio, and even if it´s close, it probably won´t ever deactivate. I was mainly talking about playback without in any way altering the signal, but reclock doesn´t exactly categorize as a "playback" application, because it also changes the audio in some cases (or in all cases, depends on the settings?).

88.2kHz/96kHz in that case.

I´m not that familiar with reclock, since I never really felt the need for it. I´m fine with my native 23,976p, 24p, 47,8p, 48p, 50p, 59,9p and 60p modes on my Eizo CG243W. I simply cannot see the extremely small deviations, even if I know that they are there.

[Qaq]

Most sound devices have x48 clock, so I guess 96 will be optimal value.

[e-t172]

http://imgs.xkcd.com/comics/duty_calls.png ( http://xkcd.com/386/)

Well, resampling to the same sample rate as the source will butcher the trebles IME/IMHO. It's measurable too: http://forum.slysoft.com/showpost.php?p=227357&postcount=3996

I never said it was not measurable. It is, absolutely. But just because you can measure it doesn't mean you can hear it. In fact, according to your own measurements, it is indeed not audible: your spectrum graphs are showing harmonics which are -100dB below the fundamental and your THD figures are well below 0.01%. These levels are way below what the human ear can perceive, and they are several orders of magnitude lower than the distortion generated by your speakers or headphones.

So, thanks for proving my point.

It's all discussed there really: http://www.audioholics.com/education/audio-formats-technology/upsampling-vs-oversampling-for-digital-audio

That's unrelated. Modern DACs already oversample internally, even if you feed them with a 48 kHz signal. Actually, because modern DACs use Sigma-Delta modulation for numeric to analog conversion anyway, the input sample rate is often not relevant to begin with.

Dan Lavry confirms it too: The Optimal Sample Rate for Quality Audio (From: http://www.lavryengineering.com/pdfs/lavry-white-paper-the_optimal_sample_rate_for_quality_audio.pdf)

Hum... what? I did not read the entire paper, but its conclusion it at odds with yours.

Also, jitter is usually lower at 96kHz than at 48khz: http://www.cirrus.com/en/pubs/appNote/AN339REV1.pdf

Again, irrelevant. You're quoting a paper about S/PDIF receivers. S/PDIF has documented jitter issues due to the way it is designed in that it needs clock recovery at the receiving end. This does not apply to most people who are using a PC sound chipset or an HDMI receiver. Good DACs have inaudible jitter to begin with, no matter what the sample rate is.

You really need to stop spreading such misinformation and start putting your measurements and the documents you're quoting in context before coming up with issues that don't exist.

[leeperry]

reclock doesn´t exactly categorize as a "playback" application, because it also changes the audio in some cases (or in all cases, depends on the settings?).
It does if you enable its frame rate adaptation(and it will resample 48kHz to 47952Hz if you play 23.976fps content in a 24.000Hz multiple), but this is completely irrelevant to this thread I'm afraid. There are people to whom everything sounds/looks the same(jitter is inaudible, yada yada)...be it then, lucky them. I wish it all sounded the same to me too.....I really do.

[Qaq]

Modern DACs already oversample internally, even if you feed them with a 48 kHz signal.
What for? Oversampling is useless according to this:
Use 48 kHz. Higher sample rates are snake oil and do not present any audible benefit. (http://people.xiph.org/~xiphmont/demo/neil-young.html)
Actually, because modern DACs use Sigma-Delta modulation for numeric to analog conversion anyway, the input sample rate is often not relevant to begin with.
Oh, I see...often. Can you explain to us how exactly is PCM1796 dealing with 48 for example? Quotes from datasheet are welcomed.
Good DACs have inaudible jitter to begin with, no matter what the sample rate is.
Would you mind people try different options to find out if they have *good DACs* or not? Just like madshi does with scalers.

[e-t172]

What for? Oversampling is useless according to this:

I see how it can get confusing. Here's the explanation: when I say that you don't need more than 48kHz, I mean that sample rates higher than 48kHz do not convey any additional audible information.

The reason why a DAC is oversampling is because you need a low-pass filter in the analog stage to remove the high-frequency aliasing artifacts from the Sigma-Delta demodulation. It's much easier to do that with a high sample rate because the aliasing artifacts are higher in frequency, so you don't need a steep filter to filter them out, so it's easier (and cheaper) to build.

If the DAC was working at 48kHz, it would need a filter with near-zero response at 20kHz (top of the audible range) but high attenuation above 24kHz (Nyquist frequency) which is an extremely steep analog filter and so difficult to achieve. On the other hand, if the DAC is working at a higher sample rate then it's not an issue. That's why it's always oversampling internally, no matter what sample rate you use for the input signal. But again, it's an implementation detail. It's about preventing aliasing artifacts in the DAC process, not preserving information that is not audible anyway.

As an example, a quick Google search returns an Analog Devices paper (http://www.analog.com/static/imported-files/application_notes/292524291525717245054923680458171AN283.pdf) which describes how this works (see figure 6.34, page 12). In this example there is a 8kHz input (pretty Low-fi, heh) which is oversampled to 1 MHz using a digital filter, then modulated in delta-sigma, then converted to analog, then low-passed by a simple analog filter. Here the analog filter is simple to implement because the conversion is done at 1 Mhz, which leaves plenty of headroom above the 4kHz Nyquist frequency of the original signal.

Oh, I see...often. Can you explain to us how exactly is PCM1796 dealing with 48 for example? Quotes from datasheet are welcomed.

Well, it's basically an implementation of the design I linked (see page 7 of the datasheet (http://www.ti.com/lit/ds/sles100a/sles100a.pdf)).

It's oversampling by 8x, meaning that a 48 kHz signal will get oversampled to 384 kHz by interpolating, then low-passing in the digital domain. For audio one would almost certainly use the sharp roll-off mode for the lowpass filter, so according to page 4 you get a flat response up to about 21.8 kHz, no issue here. Then it gets modulated (probably some kind of Delta-Sigma modulator), converted to analog (DAC), and then it goes through an analog filter which doesn't seem to be specified (probably because no-one cares, it's not relevant to overall performance) but I would guess it's a low-pass filter with a pass-band extending up to fS/2 (Nyquist frequency of the original input signal) and a stop-band beginning at fS*4 (Nyquist of the modulated signal).

If you're feeding it with 96 kHz, then you can basically multiply each of these numbers by 2, although I don't see the point.

[dansrfe]

Okay, some additional questions:

What is the optimal setting for "Format" in ReClock? Currently, I have it set to 24-bit padded to 32-bit because setting it to 24-bit integer results in ReClock giving me an error saying that the audio device doesn't support it even though the Realtek chip is stated to support 24-bit. Maybe that implies padding to 32-bit?

Suppose, I'm NOT using WASAPI: What should I set for the "Default Format" under the speaker's properties page in Windows? This is confusing for me because doesn't this dictate the frequency output of computer generated sounds as well? Like certain software generated tones (not pre-recordings). Should the sample rate still be 48000 Hz= here?

[Qaq]

Here's the explanation: when I say that you don't need more than 48kHz, I mean that sample rates higher than 48kHz do not convey any additional audible information.
I see. But we were talking about oversampling in ReClock and not about tracks with high sample rates.
The reason why a DAC is oversampling is because you need a low-pass filter in the analog stage to remove the high-frequency aliasing artifacts from the Sigma-Delta demodulation.
Thanks, I know it. So we have found that oversampling is a good thing and will be done anyway in the ReClock or in a DAC.
As we looking for a perfection, we can't avoid the question here: what is the difference? As we know, ReClock provides the different quality options for its own resampling. And what do we know about dac's oversampling algorithm(s)? Hmmm... Is there a chance that ReClock provides a better resampling algorithm? So we can prepare a sound in best possible quality in ReClock and then just simple:
If you're feeding it with 96 kHz, then you can basically multiply each of these numbers by 2, although I don't see the point.
What is the optimal setting for "Format" in ReClock?
The higher bitdepth that supported by audio driver, digital interface and reciever/sound card/DAC.
Currently, I have it set to 24-bit padded to 32-bit because setting it to 24-bit integer results in ReClock giving me an error saying that the audio device doesn't support it even though the Realtek chip is stated to support 24-bit. Maybe that implies padding to 32-bit?
24int and 24 bits padded to 32 are the same thing for the sound accuracy. HDMI requires 24 padded to 32.
Suppose, I'm NOT using WASAPI
You're still using WASAPI but in Shared mode. With WASAPI Shared you don't need to cut float to integer in ReClock and may output it as is in 32float to windows sound mixer so the last one could do its job and dither 32float down to whatever you set at the very final stage. What to set? Try 48 and 96 and make your pick.
The main ideas:
- ReClock and windows sound mixer operate in float, while audio driver and sound devices accepts integer.
- convertion from float to integer must be done at the very final stage to keep more accurasy.
- the higher bitdepths keep more accurasy.

[e-t172]

I see. But we were talking about oversampling in ReClock and not about tracks with high sample rates.

Indeed.

Thanks, I know it. So we have found that oversampling is a good thing and will be done anyway in the ReClock or in a DAC.
As we looking for a perfection, we can't avoid the question here: what is the difference? As we know, ReClock provides the different quality options for its own resampling. And what do we know about dac's oversampling algorithm(s)? Hmmm... Is there a chance that ReClock provides a better resampling algorithm?

The quality of the oversampling in the DAC is not relevant in itself: what's important is the overall specifications of the chip. It's easier to think of it as a black box with a datasheet than try to analyze its internal parts (which are not fully specified anyway). Looking at your DAC's datasheet, we can see that with a 48 kHz sample rate (or above), it is flat until about at least 21.7 kHz and its distortion is way below 0.01%. Unless you have additional data that indicate otherwise, in light of these numbers it is fair to say that this DAC does not alter the sound in any audible way, no matter what sample rate you feed it with. So we don't really care if ReClock's oversampling is better than the DAC's or not, as you wouldn't be able to hear the difference.

Regarding ReClock, I don't know what the quality options relate to in terms of lowpass FIR filter length, so it's difficult to answer. We would need to know the windowing function as well. I wouldn't be surprised, however, if the difference between the different quality options was extremely difficult (if not impossible) to hear, as it mainly affects the very top end of the frequency response. If you're using the best quality option then it is extremely likely that using a 96kHz output sample rate in ReClock is just a waste of CPU time. It would be possible to verify this by playing a test signal through ReClock and then intercepting it before it gets to the audio output (e.g. using Virtual Audio Cable), but I don't have time to do that.

[nevcairiel]

How i always saw upsampling with ReClock was to give it some more data to work with, because it always needs to resample anyway.

So i figured, if it needs to resample from 48000 to 48050 because of speed adjustments, why not tell it to resample to 96100 instead, and hopefully have simply more samples where it can put its data, possibly hiding some distortions (not that i'm likely to hear them)

[leeperry]

if it needs to resample from 48000 to 48050 because of speed adjustments, why not tell it to resample to 96100 instead
That's the thing, Reclock lies in its OSD....it actually downsamples 48kHz to 47.95kHz in your example.

There is no question that it's better to upsample by 99% than to downsample by 0.1%. It's measurable too ;)

[e-t172]

There is no question that it's better to upsample by 99% than to downsample by 0.1%. It's measurable too ;)

Yes it's measurable, and again, the measurements prove it's not audible. It's certainly not worth the CPU usage. It's like debating over the color of a car engine: nobody cares, it's not visible anyway.

[iSunrise]

deleted

[leeperry]

Yes it's measurable, and again, the measurements prove it's not audible. It's certainly not worth the CPU usage.
CPU usage in 2013? Are you on a Pentium 90MHz?

I use a $3K DAC, highly transparent orthodynamic headphones and I've got fairly trained ears as I've worked as an audio engineer for quite a while...so please be so kind as to not make gross generalizations of your personal experience.

I've got nothing to sell and I can't be dragged into endless and pointless internet debates, so moar power to ya if all sounds the same to your ears....total harmonic distortion measurements that prove me right don't mean anything, that makes perfect sense :cool:

[e-t172]

CPU usage in 2013? Are you on a Pentium 90MHz?

0.00024% THD? Do you have alien ears from the future?

I use a $3K DAC, highly transparent orthodynamic headphones and I've got fairly trained ears as I've worked as an audio engineer for quite a while...so please be so kind as to not make gross generalizations of your personal experience.

I've got nothing to sell and I can't be dragged into endless and pointless internet debates, so moar power to ya if all sounds the same to your ears....total harmonic distortion measurements that prove me right don't mean anything, that makes perfect sense :cool:

Okay. So I'm assuming you can produce the following:

- Measurements for your $3K DAC (seriously?) that shows that its distortion is less than the one produced by Reclock running at 48 kHz (0.00024% according to your own measurements). Good luck with that. Even professional lab equipment like the Audio Precision SYS-2722 ($25,000) is incapable of generating an analog sine wave with distortion that low.

- Documented double-blind studies that show humans are capable of detecting 0.00024% of distortion. Let me spare you the trouble: they don't exist.

When I say that it's not audible, I'm not speaking from personal experience, I'm speaking from what is widely recognized in the scientific community regarding the audibility thresholds of the human ear.

[leeperry]

I forgot to mention that Reclock's new resampler runs in MT(1 channel per thread), and you really shouldn't assume that what you hear(on some undisclosed test equipment BTW) is what everyone hears, even on mid/high-end gear. Or worse: repeat verbatim what you read on the internet(it's written so it's true), you might consider running real world tests and talk from first hand experience for a change.

The motive of those WaveSpectra measurements on the Reclock forum was to convince James into providing upsampling capabilities in Reclock, and I very much succeeded.....which is all that matters to me. Feel free to rerun then with 10/15kHz test tones and you might be in for some pretty drastic surprise. Unfortunately, I don't plan on making them myself as I luckily happen to have better things to do with my time.

You can downsample to 32kHz in the lowest quality mode for all I care personally, it all sounds the same remember? ;)

I'm unsubscribing from this thread now.

[e-t172]

I forgot to mention that Reclock's new resampler runs in MT(1 channel per thread)

What does that have to do with anything?

and you really shouldn't assume that what you hear(on some undisclosed test equipment BTW) is what everyone hears, even on mid/high-end gear

Again, I am not talking about what I hear. What I hear is irrelevant.

Or worse: repeat verbatim what you read on the internet(it's written so it's true), you might consider running real world tests and talk from first hand experience for a change.

You seem to live in a parallel universe in which scientific double-blind experiments have no value and well-documented psychological biases (http://nwavguy.blogspot.co.uk/2012/04/what-we-hear.html) regarding subjective audio evaluation don't exist.

I suggest you do the following: take some audio file that you're familiar with, apply some sample rate conversion that you consider audible (while staying close to or above 48 kHz), and then compare with the original using ABX software such as foo_abx. I am ready to bet $500 that you (or anyone else, for that matter) won't be able to pass the test. (but of course I won't, because I have no way to know that you're not cheating)

You can downsample to 32kHz in the lowest quality mode for all I care personally, it all sounds the same remember? ;)

Of course it won't sound the same. If you downsample to 32 kHz then Nyquist is 16 kHz, so you'd be removing all audible content between 16 kHz and 20 kHz (and even more than that because of the non-perfect low-pass filter). That is a valid concern. 0.00024% THD, on the other hand, is not.

I'm unsubscribing from this thread now.

It's a shame to see these superhuman alien ears go.

[welp]

Hey guys, a got a quick and hopefully easy question.

I've got a htpc set up (using hdmi) and with reclock under wasapi exclusive mode I'm not getting sound from videos with audio above 24bit. I checked my receiver spec sheet and it reads it only supports up to 24bit, so I'm guessing this would be why. What I need to know is how I can get around it or if I should just switch to wasapi shared mode.

[Qaq]

Like I said above, with HDMI you have to set "24 bits padded to 32". Please try to read before posting.

[welp]

PCM output under reclock is already set to 24 int padded. I've tried a ton of options already and am just stuck. I could install and configure ffdshow to handle only the specific codecs from the videos giving me trouble but I'm looking for a cleaner solution. The exact error getting thrown when trying to play is "the audio format is not supported by the hardware /driver" I only get it with videos using 32bit PCM. Anyways that's why I'm thinking it's my receiver, let me know if you can think of what I'm doing wrong.

[Qaq]

Try to specify the sample rate. Note Nvidia HDMI doesn't support 88,1kHz and 176,2 for example.

[welp]

Thank you Qaq, I knew it was something simple I missed. This whole thread has been about sampling rates I should have messed around with it a little. After reading everything I just didn't think it would matter and kept it at default.