Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards Any Kawai piano with USB audio interface?

ever since they got bought up and passed around, their entry and middle products are really bad. same with tannoy.

It's a common misconception that manipulating digital signals is a perfect process that can't possibly result in any signal degradation. You can see how this is wrong pretty easily by doing some common things to a digital photograph in Photoshop. Suddenly color gradients that were smooth have noticeable banding, for example, or you lose shadow or highlight detail.

You also can't assume that any device that manipulates signals digitally will end up with exactly the same result with the same quality. I would trust a Universal Audio interface to mix and manipulate digital audio with better quality than whatever a digital piano maker sticks in their keyboard, for example. On a piano, mixing in external audio is much more about convenience than quality.

The real reason for 24- or 32-bit digital audio at 88Khz or 176Khz isn't because it sounds any better than 44.1Khz but because it allows you to capture, mix, or otherwise manipulate (e.g. EQ) digital audio and maintain enough precision that you can convert the final result back to 16-bit 44.1Khz without the signal being a mess. Similarly for working with digital photographs at higher resolutions and color bit-depths than you use for the final output.

the bit depth helps with any operations that require dithering.

certain operations in photoshop do not dither, so if you're working in 8bit, there is a higher likelihood that you get banding.

also, on nvidia their color correction does not work properly in the non-pro cards, you have to use a registry hack to get their dithering engine to work right. only their pro cards dither properly with color correction. amd cards dither properly with all color profiles at all bit depths.

Dithering is an example of how manipulating data digitally is not a perfect process that always maintains the quality of the original. Dithering is an algorithm that makes up data out of thin air to approximate a result, with different algorithms producing different results of different quality.

you can think of bitdepth as an attempt to preserve high frequency textures. for photoshop they avoid dithering in certain operations because if they did, the end result would be too smeared. so they leave it out, until you do the final composite and manually decide where to use decontouring.

The addition is a linear operation. So I don't see a misconception here. Any affine transformation is not going to introduce loss in the signal.


You need to be specific. This can be inherent or an artifact of certain down/up-sampling.



There is no signal modulation or multiplication when it comes to adding two waves. What you are saying can happen if DSPs or EQs are applied. If the signal is bypassed, it is just 0s and 1s that are added together.

The only scenario that signal deterioration can happen is that the DACs are different or the host hardware has lower precision. In the case of audio interfaces, you always know about the sampling rate and depth.


I agree with this and I mainly avoid using onboard I/Os of my keyboards because they are the bottom-line hardware.


Signal quantization (what you are referring to here) can happen for sure, but look at it this way:

How is your internal interface is going to add two digital signals together? It will be almost 99.99% identical to your external interface. The same error in signal quantization after addition will happen on both platforms.

So not much gain or loss here if you mix digital signals on your keyboard or in your hardware as they both suffer unless you know the details of each DAC... and it probably is so small and insignificant that you need some software to analyze the signal to tell the difference.

I purchased a mixer (XR18 to be specific from Behringer) and it is quite the reverse. Music Group actually is better than it was before as I own an old pair of Behringer monitor speakers that are absolute crap.

These tiny little devices from Behringer outperform most of Yamaha's entry-mid-level gears. One example is the Steinberg UR series which is the bottom line when it comes to pre-amp quality.

Also, these XR18s are amazing little mixers. The internal wifi is junk but the rest of it is amazing. They are so good that the entire industry started to copy them. But no one has managed to outperform XR18s in price-performance departments.

18 digital I/O.
Good preamps
4 shared effects
The compressor on each channel, EQ and so much more.

There isn't an alternative in the market other than MIDAS which is the same concept with better preamps.

When you start adding signals then you can overflow/clip the signal and then you have to start doing other manipulations to fit the signal into the format. Those other manipulations can negatively impact the quality. When you're mixing you're also not just adding but also multiplying or dividing in many cases in order to balance the relative levels of the signals in the mix. So you might be amplifying noise or magnifying the lower resolution of a low-level signal when you multiply it in order to mix it with a higher-level signal for example. Or if you have to divide a high-level signal so that it doesn't overwhelm a low-level signal then you might be throwing away some data (reducing resolution and/or dynamic range) to squeeze it into a smaller range.


Let's say you have a smooth color gradient with color values of 1, 2, 3, 4, ... within a given RGB channel. Then you decide you want to adjust the brightness and/or contrast of the image. Suddenly you end up with values with gaps in them that are no longer smooth, and you can detect jumps (banding) in color rather than the colors being smooth. That's when you have to start applying algorithms of varying quality which literally fabricate data to try and make the quality loss less noticeable based on human perception (to an alien you might be making it look worse).


As I said, even adding signals can force you to do other types of manipulations. But all these devices are also doing EQ, reverb, etc. which obviously varies tremendously based on the quality of the algorithms being used.

You do normalize the signal to fit in the boundaries (as I mentioned in my original post). It is not going to change anything. This happens in both the digital and analog worlds btw.

If it is digital, you need to normalize it (easily). If it is analog, your signal will be too hot to handle and either blow the hardware on the other end, or it will be distorted. You need to adjust the gain in this case.

In the live performance scenario where we use analog signals, we leave some headroom for each input. We use compressors, limiters, etc on analog signals to set a boundary so that "when they get mixed together this won't happen".

If my signals are fed or always adjusted with gains I can always take care of this issue and the argument is practically really irrelevant.

Like I said, and I am saying it again, it doesn't matter what setup you are using this will happen on both internal and external interfaces in the worst-case scenario if the user doesn't know a thing about mixing signals.


What are you trying to prove here?


It is impossible to avoid it if it happens. Do you know a method to mix two signals that don't manipulate signals?



The multiplication and additions are all affine operators. Which means they are linear. In terms of information loss (I mean statistically speaking) if you "increase" the amplitude(or energy of a signal let's say), you will lose information. If you decrease the altitude, you will not lose information.

In fact, the noise levels will be less in the reduced signal (downsampling).

I still struggle to understand what you are trying to achieve here?



The operations are the addition and power (usually you reduce the signal so less than 1 power). If you increase the gain you raise the signal to the power of 10 (let's say). Therefore you'll magnify the noise levels but this is for the scenario when you have a static recorded signal. When you have your keyboard, it has been adjusted to fit within the bit depth of the recording and you can always set the gains to -0- on both ends and adjust them while you're mixing them to minimize the loss.

Again, if you are talking about quantization errors (which is irrelevant to be honest with you because it happens all the time and everywhere) and you are saying this produces an inferior signal, then you are more than welcome to "invent" your own method that doesn't suffer from these artifacts.


There is no smoothness in your example. 1,2,3,... are discrete values.



That's because of 3 reasons:
1- The most important one: your original input is discrete, and is not smooth.
2- Your algorithm can interpolate the in-between values but it doesn't.
3- Your graphics card and your monitor's bit depth settings are incapable of displaying the colors. You can purchase true 10bit or even higher bit depth monitors at some astronomical prices.


When mixing, everything happens live and you can adjust everything. In the case of an offline process, the result is the same (if not better) on an external device.

So I am really confused at this stage about where we are heading from here?!



If you read my initial post, you can by-pass EQ, reverb, and everything else.

On both of my MOTIF, and Behringer X18 I can bypass everything. The RAW signal can be sent out easily.

It was used as an example of how simple a mixer can be-- the minimal device that qualifies as a mixer, not as an example of audio quality, although it is so simple that its audio quality will be perfectly fine for many applications.

Digital noise differs from analog noise, but it exists. Digital noise results from roundoff errors. When you add two (digital) quanta, there can be floating point roundoff error. Therefore, a digital mixer is not noise-free.

There also is quantization error when an ADC has to round off a sample to the number of bits in the sample quantum (16 bits for Kawai DPs).

They wouldn't. But not everyone has a studio where the DP is the central focus and everything else is viewed as a peripheral to the DP. For many people, the DP is just one device. Mine has audio cables leading from the AUDIO OUT of the DP going to an analog mixer, and a midi cable leading into the DP so that its non-piano sounds may be played with a semi-weighted action. Midi over USB would not even work for that because the cable is longer than the 16-foot maximum for USB 2.x so that the cable runs along 3 walls instead of across the middle of the floor.

Spend some time considering how people with 88-key midi controllers manage to use their product without this feature (typically without any DAC or speakers built-in), and it may help clarify the issue for you. (Akai had a controller with a multiplexed USB connection and built-in DAC, but they recalled it due to some lubricant issue, and as far as I know, never continued selling it).

The sample rate just has to meet or exceed the Nyquist frequency and quantization error from round-off will be the digitization error of concern.

True. Twice the Nyquist rate will give you a "theoretical" similar recording quality.

I have mentioned this before and it wasn't welcomed but I really do prefer digital out over USB audio. I wish Kawai could at least give us this.

Sorry, I now see that the diagram I've posted has already caused miscomprehension because I've only shown the bit rate diagram in the absense of the sample rate diagram.

To clarify for people not familiar with what's being talked about...

The sample rate diagram is below:

Nyquist is about what frequencies you can capture. Quantization error is about how many bits of resolution you use to encode the amplitude of the signal. If you sampled at 44.1Khz, with 1 bit per sample, for example, you would only be able to record "sound" or "no sound" for each sample and not any specific volume levels in between.

16-bit quantization gives you a lot more possibilities, but now let's say you record a very quiet source that only uses the lower part of the range and you mix it with something louder where you need to boost the volume of the quiet source. Now you've magnified the quantization error. This is one reason 24 bits or more are often used to capture and process digital audio and only once it's been fully processed does it get converted to 16 bits. But if you are mixing 16-bit sources inside of a digital piano then there's a lot more possibility for quantization error to be magnified.

They do-- via recording to USB memory.

I think Abdol means something a little more real-time, like the coaxial digital out on the V-Piano

There seems to be two camps in this conversation.
1. Those who have a keyboard that has inbuilt USB audio and are delighted with the simplicity and convenience of single cord connection for VST's and digital audio in, and
2. Those without inbuilt USB audio who are desperately trying to justify why its not needed.
Human nature at its best!