Piano World Home Page Forums Our Most Popular Forums Digital Pianos - Electronic Pianos - Synths & Keyboards The DPBSD Project!

@dewster, thanks for the analysis of CA-63! Some comments:

the mosquito Noise might be due to the fact that I did an MP3 compression two Times which is not really a good thing to do. I recorded the sample directly on a USB stick using the internal Recording functionality of the DP. Unfortunately wav Recording was not possible, it always led to jumps and crackles in the Recording (another flaw of CA-63). So I imported the MP3 into audacity and normalized the file since it was quite low volume and then recompressed it.

Interesting that there is no obvious resonace... This was activated and clearly audible. Maybe when playing something through external MIDI still deactivates any resonance effects? Okay, another flaw! There are plenty of (rather small) defects that should be resolved in future Firmware releases. The Recording function is very sluggish and regurlaly freezed the DP so that I needed to reset the DP... Need to talk to KAWAI James soon!

I was seriously considering the CP1 until I found out the massive dim-able product logo on the back only has three levels of illumination. At that price-point I expect at least four levels. Yamaha blew it right there IMO.

Sorry if this has been addressed earlier (didn't feel like looking through all 9 pages) and I'm a bit late to the party here, but has anyone (dewster) compared a raw wav to the corresponding mp3 to make sure that the encoding itself isn't introducing artifacts that are then being measured. I would think for this to really be useful, everyone would use raw wav's thereby eliminating the quality of the encoding codec if this is to be something truly scientific or analytic.

Like I said, if this has already been addressed just point me to the post(s) and I'll shut up

Maybe your memory stick is to slow to keep up.
You might try using a card reader and high speed SD card or even maybe a USB hard drive.

To record a wav file the DP must send about 180,000 bytes per second to the memory stick. That does not sound like much but some of those memory sticks are very low performance devices.

And you say you aren't biased in these tests...

So much for legitimacy.

Snazzy

I don't think a modeled piano would be the best point of comparison. The modeling algorithm may have step points or look-up tables, the result of which could mimic the spectral "finger printing" of sample stretching. Better to use a 100% guaranteed unstretched real piano sample instead.

When I see very little note stretching in a DP, particularly when it is confined to the upper end of the keyboard, my inclination is to believe that the actual piano itself had some less-than-perfect sounding notes, and stretching in that case to remove them is more of a blessing than a curse. This is of course pure conjecture on my part, but it does make a certain amount of sense.

I'd be very interested in any comments from your engineers James.

That was sarcasm, BTW. Guess I should have used a smiley or something. It's getting so you have to explain even feeble attempts at humor to death around here.

Dewster: Please consider expanding your test to include repitition rates of the actions. To do this properly I guess you'd need a pretty sophisticated device that could strike the keys with a variable rate. (and probably variable velocity and variable strike depth as well)

One thing I'm interested in is whether any high end pianos WITHOUT triple-sensors can outperform something like a Casio PX-130, which is very low cost, but yet does have a triple-sensor action, for example.

So get onto this ASAP please - chop chop.... ;^)


Greg.

Good point.

If this where a purely listening test I'd say it does not matter. But seeing as the test is analytic, it might.

For example does the MP3 encoding process preserve phase information at all frequencies? Is amplitude quantitization uniform across all frequencies? MP3 is designed to preserve only what the human ears would notice and remove the rest.

Perhap MP3 encoding artifacts could explain the apearence of note stretching inthe upper regiters of the new Kawai. The MP3 encoder simply removed whatever made the notes different. Certainly the MP3 encoder effects the noise floor (All compression introduces noise)

If you want to post the original MP3 I'll take a listen to that too.

You want the keyboard testing department, that's down the hall.

Thanks - I found the right department: ;^)
http://www.youtube.com/watch?v=muCPjK4nGY4
(warning: off topic. This is the "speaking piano" clip - looks like they definitely have the right kind of hardware to do the reptition test, though!)

Greg.

Yes, I agree, good point. I did actually wonder about this myself. But most DP compression artifacting is so crude it is clearly audible, and this correlates pretty well with what I am seeing in the MP3s. I also think a WAV file would be rather onerous for people to transfer and store, so if MP3s are sufficient, then they have a clear advantage there.

I will analyze a WAV file source and an MP3 generated from it to see if it makes any difference and report back.

Chris, one (potential) major issue is that there really is no standard for the actual lossy compression part of the mpeg3 "standard", so in theory seeing the varying "quality" on these graphs could be at least in part related to different codecs being used. From a practical standpoint, this may not be an issue if there is general industry adoption of a few (or one) codec, but I don't know if this is the case. And to answer your question, I believe that the quantization values vary depending on frequency (and even then isn't necessarily constant for a given frequency, you can also look for the presence of other frequencies at a given point and assign bits as necessary).

dewster, I understand that it is far easier and faster to upload mp3's, but if there are artifacts being introduced by the compression, then a significant part of what you are attempting to do may be rendered moot. Better to get this part squared away regardless of any implications to convenience.

Wow, another 2010 flagship, thanks loads!

Your file is quite something. It's hard to tell what's going on in there. I think I can hear something like looping, but not quite, and it isn't obvious in the spectral views. This won't be a quick or easy review, I'll have to devote some time to it.

kawaian, this thread perhaps isn't the best place to discuss these points. However, since you have raised the matter, please allow me to respond.



I do not believe the CA93/CA63 currently allows MIDI files stored on USB memory to be loaded into memory, played back, and simultaneously resaved as WAV/MP3 audio files. Therefore am I correct in assuming that you used a computer to play back the MIDI file?



As ChrisA correctly notes, if the CA93/CA63 is able to record to MP3 without any problems, but WAV files result in pops and crackles, it's most likely that the transfer speed of your USB stick is too slow. This point is highlighted in the 'Troubleshooting' table on page pp.82 of the owner's manual:

The file transfer speed of the USB memory device may be too slow to
play the audio file. Please try using a different USB memory device,
ensuring that it conforms to USB2.0 Hi-Speed standards.



Again, this is likely due to a slow USB stick.



Indeed, please feel free to send me a message.

Kind regards,
James
x

Thanks! I probably get to it tomorrow.

There's a lot more to it than simple convenience. As I said above, what I see in the MP3 files correlates closely to what I can hear, so I've come to trust it to some degree. And many of these things are very clear, both audibly and visually, so there is no way I'm hearing things or hallucinating with those. The Kawai CA-63 stretching issue is near the limits of what I'm comfortable including in a review, and KAWAI James' feedback there could be very helpful in resolving that one way or another.

And as I said above I will do an in-depth comparison to see how a WAV and MP3 from the same source compare, though you may want to do this experiment yourself. Over at the Adobe site you can register for free and then download a free 30 day demo of Audition. Please don't take my word for anything, that's not what this is about at all. If you do get Audition installed, please download some of the MP3s and see how your analysis compares to mine - I'd be glad to have some feedback, even if it is negative.

Anyone here done graduate work in psycho-acoustics or audio compression want to chime in? I don't want this effort to boil down to whether people trust me or my expertise (such as it is) or not.

A read an interview with one of the Roland engineers. This has before Supernatural had been applied to acoustic pianos but he describe the process as sample based but something different from what Yamaha and other do. They have multiple channels of samples for each note and what we hear is the weighted sum of the channels where the sums are envelopes.

Some channels might contain overtones and others a fundamental. some channels might be note stretched and others not. Some may be seconds long and some only contain note attack. And then they say the sample is really a composite from several acoustic instruments. But then when I read this they were doing drums and brass instruments, not ACs and they said the method differers for each instrument.

Any link to that interview? I'd really like to read it.