Well, one would expect that the absolute top of the line acoustic replacement costing between 2 to 5 times as much would have more of a what it takes to succeed as a solo piano instrument than the CPx line which is often going to be played in a combo perhaps more often than not with electronic piano sounds or AP sounds that have been distorted with all the adjustment capabilities.

+1

While the attack sample is longer in the N3, the looping in the CP1 is freer of looping artifacts. I just took another look at the CP1 and Yamaha appears to be using a rather static set of tones for the decay, perhaps a highly processed loop of some sort.

If they absolutely must loop, they either need a longer loop or a more processed loop IMO.

N3 sample memory size calculations:

- Sample lengths are (C2:C9): 4.4,4.3,3.9,3.5,3.0,2.1,?,? seconds.
- Loop lengths are (C2:C9): 0.7,0.67,0.67,0.67,0.56,0.43,?,? seconds.

Add sample and loop lengths together, assume the unknown highest sample|loop lengths are 2.1|0.43 seconds. This gives:

(25.4 + 4.56) / 9 = 3.33 seconds avg per sample

The big assumption is how many velocity layers there are. Let's make a reasonable guess and say 6. So:

3.33 sec/note * 6 (layers) * 2 (stereo) * 2 bytes/sample * 44100 samples/sec * 88 notes = ~310MB

Not too shabby as these things go. Buying on the web from Avnet, a 1 Gb part costs $16.50 (qty 1 pricing). 1 Gb = 125 MB; 310/125=~3, and $16.5*3=~$50. I'd be surprised if Yamaha spends more than 20% of this, and $50*0.2=$10.

If the M3 costs $10k (?), that's around 0.1% of the entire retail cost to store the main piano sound sample set. With 200%(?) distribution markup, that's around 0.4%.

I agree. That the Avant Grand "out-performs" the CP1 in terms of its technical spec surprises me. I would imagine though, leaving aside the AG's tactile response features, that subjectively the CP1 sounds better/nicer. I'm probably over-simplifying this but if Yamaha combined the AG's sample lengths and absence of stretching with SCM they would have a compelling product both technically and subjectively (leaving aside issues of personal taste etc).

Maybe we are pretty close to something groundbreaking for a hardware product...the Roland Supernatural stuff and Yamaha's SCM are moving the game along significantly whilst still using samples as a basis for tone generation.

Interesting stuff.

Steve

These techniques are interesting, and SN in particular has nice note decays. I would still highly prefer they stop playing games and just give me a whole real sample. If they want me to pay $50/GB - even though the street price of Flash is more like $1 - I'll do it. Hell, I'll pay $100/GB.

Dewster. What is the randon access speed for thr flash ram you quoted? Access speed controls latency.

Also did you accouunt for the four channel sound? I could have missed it.

Also if the velocity layers are blended then you need to read out two layers so that you can interpolate. With four channels that means eight 44.1Khz simultaneous 16-bit streams. If the N3 allows layered sounds then more (16?) simultaneous streams all with sub-millisecond latency.

I'm trying to figure out how the ram must be organized. Certainly not as a linear array with 16-bit at a time readout. It must use some complex bank switching. This means many smaller parts and higher cost.

Computer based samples can be stored dynamic RAM which has nano-second level access time to every word but Flash has to cycle to the correct place before you can sequentially read the data. I bet they use a good size stack of banks or maybe they copy the flash into dynamic RAM when you select a voice?

If it were me designing this I'd use a dynamic RAM that was loaded when you select the voice(s) but only because I can't design a complex multibank switch. We don't know how yamaha designed it.

There are other cost drivers. How much engineering time was spent and how many N3s do they sell? In the extream case Oracle sells you software and for $50,000 all you get is an e-mail with a license key. What does an e-mail cost to send? No, you are paying for Oracles 5,000 person engineering staff and a relative small customer base. I won't even tell you what my customer has to spend. We have a user base on just one and he pays for half a building full of people, for years and years...

Don't misinterpret. I'm just trying to understand what is really inside. I doubt it work like an iPod but your cost analysis assumes an iPod-like design

it wouldn't be that difficult to have a DP that during startup would copy the full sample set to DRAM and use it from there...

It is quite surprising to see general-purpose computers leading the way compared to dedicated hardware (in both sampled and modeled pianos, although the vpiano seems to have caught up modeling-wise for now) but then again the market for DPs is tiny compared to the software market, and also composed in significant % by people that want to buy a piano and play it for years as-is

I agree about the delay in voice switching if flash were copied to RAM. I don't notice such delay so I doubt they do it that way.

One good clue is that DPs all have fixed maximum polyphony. This implies that the ROM has 128 parallel readout channels if the polyphony is 128.

So I take back what I wrote above about needing 16 parallel channels, no you need 128 simultaneous streams of 16-bit samples at 44.1Khz. or and aggregate rate of 11.3 MB per second. But then layer interpolation doubles this 256 streams and 22.6 MB/sec. Polyphony serious complicates the design.

The specs look impressive when you work out what it takes to play 128 layer interpolated notes at once with near zero latency

Actually I doubt Flash is used. I bet it is a masked ROM.

are DPs you use instant-on? I thought current DP took a few seconds to come up when first turned on



or it could be that whichever asic they do to do the hardware mixing of the samples can only do 128 voices in parallel, you never know.



22.6MB/sec seems extremely paltry by today's standards honestly



I would be surprised if the latency was 'zero', not sure how this could be measured (this being a closed system, where you could also fudge the sensor location to have the 'on' sent before the key reaches its stop to make it seem the sound is output right as the key hits it) but I'd be really surprised if it was less than 5-6ms, if not more

Its not stereo, its 4 channel = 620 mb.

Yes, thanks, I didn't take that into account.

Casio uses Spansion flash in their PX-330.

Okay, what am I missing. I go to the site and I see the midi file and the readme and all the sample MP3 files and folders...but I can't see anywhere to upload my MP3 test file to. Did something change that was mentioned in the middle of the 50 pages in this thread?

Goofball Jones, you just upload your mp3 file to the default MyFiles location of the mediafire website (or your folder if you register), then PM Dewster the share link to where you file is, and he'll pull it over to his folder. You just can't upload directly to his folder for some reason.

Sorry, I should put that in the readme. MediaFire doesn't allow uploads to my sharepoint other than by me. You can make your own MediaFire account though and upload it there, it's pretty easy. There are other free shareable/linkable web storage solutions out there too.

Which DP in particular are you uploading a DPBSD MP3 for?

No problem. I uploaded it to MediaFire. Here's the link:

Yamaha S90xs Natural Grand S6. This is the S6 piano that's new to the S90xs/S70xs.

http://www.mediafire.com/file/jytnvdmewam/dp_bsd_v1.5_yamaha_s90xs_natural_grand_s6.mp3

Recorded directly from the board unto a USB drive as a WAV file, then converted to MP3 via Audacity. No modifications other than amplifying it up a bit to -1dB. This is the stock 001 piano on the S90xs, factory set-up.

Got it! The levels look good (-73dB noise floor, -1dB peak) & I see stretching - I'll review it shortly. Thanks!!

From the Yamaha web site for the S70 XS / S90 XS (they actually mention ROM size, unusual) - I wonder if the S6 sample is from the same session as the S6 in the CP1/5?:

Sounds That Satisfy
The best musical instruments start with top-quality sounds, and the S70 XS/S90 XS delivers all that — and more. The S Series features all the Voices, Performances, arps and Virtual Circuit Modeling effects found on the flagship MOTIF XS Music Production Synthesizer. Plus we've added a huge 142 megabytes dedicated exclusively to high-quality piano waveforms, for a total of 456 MB of instrument samples.

Foremost among these new piano sounds are the new samples taken from Yamaha's world-renowned S6 concert grand piano. These fully capture the richly textured sound of that wonderfully expressive instrument, and provide a warm concert grand piano sound unavailable on any other keyboard. For variation, the MOTIF XS piano based on a Yamaha CFIIIS is also available.

From what I understand, it is from the same sampling sessions as the CP1/CP5. There is a video on Keyboard Magazine's site that shows a demo of the CP1 and it was mentioned there I believe, but I'd have to go back and watch it again.

Also, you said that you see stretching...is that good or bad that you saw that right away?

Obviously bad since it is a sound artefact.

But it only matters if you can hear it.

Yes, but he seemed kind of excited to see stretching. At least that's how it came across to me.

Maybe because it confirms his valuation of DPs as mostly crappy sounding...

Ah...

Why on earth did I contribute to this then!

Goofball, I'm hearing something like reverb in the file - right at the very beginning where the level setting notes play. Do you know if there is any way to turn that off?

Because it's helpful to see how the subjective sound of a piano is build on a technical base, at least for some aspects of the sound. It can help you to focus on parts of the sound to check if you can hear artefacts or some other things that might annoy you after some time. So this is a good resource for people who intend to buy a DP. But nevertheless, it can never replace playing several DPs by yourself. That should be the ultimate decision base for buying a DP.

Oh, not too excited - just giving you some info from what I saw in the preliminary inspection. It also fails the silent note replay test. The decays are nice and long.

The reverby thing is making it difficult to assess sympathetic resonance though.

Not sure...that's the standard stock patch after a reset to factory settings. It's the piano you get when you first turn on the machine.

Having said that, I'll try to see what I can do, but it would entail editing the patch.

Hey dewster,

Here's Galaxy's newest piano, the "Vintage D":

http://www.mediafire.com/file/z41jjanzjmn/dp_bsd_v1.5_galaxy_vintage_d.mp3

I got it the other day and am enjoying it a lot. To my ears, it's a very nice sounding piano. I used the default patch which already had the Release Samples and "Overtones" (Sympathetic String Resonance) turned on. I did turn on the Damper noises but didn't change any levels. No other settings were changed so this is the "out-of-the-box" sound.

If you can disable just the reverb that would be great.

Nice levels and I don't hear any obvious reverb, so that's good. No looping or stretching, though it fails the pedal down silent replay test. I see at least 7 or 8 velocity layers in there, nicely blended over the midrange, but I can hear steps at the low and high ends. I'll get on it today.

Thanks much!!