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

Yes, very good points Andy...however, it it is always good to see both sides of the picture.

Snazzy

Wow, outside of PC samplers and modelers, this is technically the best DP I've tested.

I just added a line to the DPBSD readme file instructing people to not normalize MP3s as it seems to be introducing mosquito noise. Do you have the original un-normalized file? Also, was the reverb off (I couldn't quite tell)? Also, are there any key-up or pedal up/down samples or sympathetic resonance not captured by the MP3? It seems to respond slightly to partial pedaling, but I need to modify the DPBSD MIDI file to include a 111 pedal level to better capture this across all DPs.

Anything I should know about the setup you used to capture the recording?

Really fantastic, thanks again for that file zaba19!


-----------------
- Roland HP-307 -
-----------------
FILE & SETUP:
- dp_bsd_v1.3_roland_hp-307.mp3
PROS:
- Beautiful long natural-sounding note decay (decay times on the order of Pianoteq).
- Large dynamic range (~47dB, vel=1:127).
- No audible looping.
- Visually, the notes look similar to conventional length attack samples seamlessly blended with something like long loop samples. Not clear what the process is.
- No visible or audible stretching, notes look random in the wave and phase views.
- No visible or audible velocity switching.
- Very smoothly blended timbre variation with velocity.
- Responds to partial pedaling (the DPBSD MIDI file needs a 111 level pedal).
CONS:
- No obvious sympathetic resonance.
- No obvious pedal up/down or key up samples.
OTHER:
- Probably good enough to realistically record solo.
- Sounds a bit like the reverb isn't turned off?
- MP3 levels excellent: peak @ -3.4dB, noise floor @ -81dB.
- A bit of mosquito noise in MP3 (probably due to normalization).
- Date reviewed: 2010-02-17

Speaking of having the full picture, I'd like to know if anyone has financial incentives (from, oh, I don't know... Yamaha) to demean and dissuade dewster. It sure seems that way.

dewster is the new Galileo...

I don't contribute to this board very often any more (more of a lurker these days) but after reading this thread I thought now might be a good time to add somethign useful. There is a fair amount of FUD floating on this thread from both camps so I thought I might contribute some technical information. There IS merit to what dewster is trying to show, BUT I think there needs to be some context. He is absolutely correct that all ROMplers (DPs are a type of ROMpler with the exception of the V-Piano) use looping and stretching as a form of data compression and will continue to do so into the forseable future.

This leads to a few of questions which when answered should help people get some perspective:

1) Why in this day & age of advanced computing and inexpensive memory are they (the manufacturers) using compression at all?

First I think that everyone would agree that the customer requirements of DPs (especially stage versions) are especially stringent in the areas of noise, weight, durability and cost. For this reason the inexpensive hardware that we've all become so accustomed with in our home and business computing environments is not what's used in DPs, or pretty much any other industrial application. The ASICS, PGAs, FPGAs and their RAM are generally purpose built (hardened) and tend to be very good at one thing. They run at slower clock speeds but perform their specific functions very quickly since they're not designed for general computing. They typically run cooler than general purpose computing devices which eliminating the need for noisy secondary cooling systems like you'd find in your PC or MAC.They have VERY low power requirements which simplifies power supply designs thereby reducing weight and complexity. They are much less sensitive to electrical interference, power fluctuations, noise and surges (especially during power on and off).The choice to use this type of computing technology as opposed to general purpose computing all comes down to the design considerations necessary to meet the initial customer requirements.

Unfortunately this equipment has some drawbacks, not the least of which is the amount of addressable memory with ranges typically in the kilo to mega bytes (yes I said Kilo to Mega Bytes) and not the Gigabytes we thing of when when buying a home computer. Just like any other computer, the operating system (albeit a very simple one in a DP) must also fit in that addressable memory along with any data (samples in the case of a ROMpler like a DP). Just to make matters worse, this is not the same memory you buy at your local computer store and it is MUCH more expensive given the limited applications.

So if this computing technology is so restrictive when it comes to memory, then why use it,..... simple; it is the best for meeting the basic customer requirements (noise, weight, durability, and cost). If you were to try and build a DP using COTS (commercial of the shelf) "consumer" computer equipment, meeting the first 3 requirements would definately result in a violation of the 4th. My (very educated guess) at building a COTS consumer type computer that is silent, light weight & small enough to fit within a DP case, and meets the mechanical and electronic durability requirements of a stage piano would cost upwards of $8-$10K to the end user, "if it could be done at all"....... Oh and I should point out that a Laptop does NOT meet these requirements.

2) OK, so if the computing equipment limits the amount of addressable memory so they have to use compression; Why not use algorithmic compression to preserve datails (less lossy) instead of stretching and looping (more loss)?

Algorithmic compression is extremely computationally intensive and non-detirministic resulting in the potential of latency and increased complexity. It is possible to add more hardware to solve the problems but it requires 2 layers of storage; dedicated ASICs to handle the memory management and decompression and a much larger data bus/bandwidth all of which increase cost significantly.


3) Why are there so few layers; why not sample all the potential layers, or more accurately, why not sample enough layers so that no one could tell the difference?

Each sample eats up precious storage and quite frankly, more just isn't needed for the typical application. Most people can't hear more than 3 or 4 distinct timbre changes, or more accurately, the timbre changes or so subtle that "most" people (the bulk of the audience for this class of product) just don't care. In a mix it is quite frankly impossible to tell how many layers are used.


4) If they have to use S&L compression why are the intial samples so short and why do the lengths vary from one sample to the next?

For anyone who has done sampling, the answer to this one is obvious. It's all about finding usable loop points. Looping is an extremely difficult process (there are very few tools to help) that requires that the tail of a loop SEEMLESSLY fit into the head in a way that makes it very difficult for the average person to perceive. You need to keep the main sample since this is what gives the note its character but it's length/duration is more detiremined by the start of the loop segment than anything else which is why some intial samples are a few seconds while other are less than a second,.... it's all about finding the best initial loop point. The loop segments length is detirmined by finding a point that ends at exactly the same amplitude as the head of the loop and is moving in the same amplitude direction. Loop tend to be short so its decay is controlled by an amplitude envelope.

5) If looping must be used then why do they have the loops decay more quickly than on a real accoustic piano?

It comes down to realism in the loops and controling polyphony/note stealing. While looping, when done well (and there are a lot of examples of it being done VERY poorly) and kept relatively short isn't obvious to most but the most critical listener, if it continues for too long, it becomes grating on the ears and seems synthetic/unnatural. Also, as long as a note is sustained, it eats up an oscillator (stereo samples eat 2 oscillators). The quicker a note drops to zero amplitude, the quicker those oscillators are freed up to play other samples/notes. Once the DP runs out of oscillators, then notes must be dropped in favour of the new notes played (this is called note stealling). With newer boards having 128 oscillators (128 note polyphony is the marketing terminology but is in fact 64 note polyphony if stereo samples are used which is the norm these days), the loops can be relatively long before note stealing happens, but definately not the length of an accoustic. I would like to point out there there is very little repertoire that would require a voice/note to decay to zero naturally so the only circumstance that this normally happens is if you put a brick on the damper pedal (which is poor technique for most music except new age). In other words, having a shorter decay just doesn't matter most of the time (unless you are performing contrived tests to find it).

There is no value in complaining that DPs don't use different hardware than they do because the hardware choice is the correct one to meet the primary requirements. Yamaha, Roland, Korg, Kurz, Nord, etc are not trying to rip anyone off, but are trying to build the best products to meet the largest segment of their market. A consumer who falls out of the norm of that market can choose to build their own DP using general purpose computing, software samplers and uncompressed sample libraries, their own sound reinforcement (monitors/amps), a controller (MIDI keyboard), and can even build a case to put it all in but it will NOT be suitable for general use.

DISCLAIMER: The following is my opinion!!

I have sevaral computers dedicated to composition, sampling, and sound production and I have just about every piano sample library that's been published (my son composes for film, television and games so we have most of the big orchestral libraries as well in our studio); I also have several DPs, synths, and workstations so I think you can understand that my opinion comes with some experience.

If we are rendering a solo piano performance, then I will ALWAYS use a sample library (for many reasons but the obvious one is the depth of realism attained in the recording). If I'm rendering an ensemble or orchestral piece, then I will opt for either a smaller sample library or one of the better DPs/keyboards/sound modules. If we are rendering a POP (guitar, drums, keyboard) ensenble piece, I will almost always use a keyoard because that is the correct sound. If I'm just playing piano, I rarely use a sample library becasue I have yet to find a library that doesn't produce a feeling of being disconnected from the instrument,.... Keyboards/DPs trounce sample libraries for that feeling of being connected to the instrument and the better instruments are "Good Enough" in their realism to not leave you feeling like you are playing anything other than a Piano. I also have an acoustic grand which is also enjoyed on a regular basis but personally I actually prefer to play (recreationally) the digital instruments for their variety of tone and consistancy of tuning and touch. No one in the family is a concert pianist (although my son makes his living at music production) so having infinate variety of timbre, or resonance, etc is meaningless to me.


There's my 2 cents,.... take it for exactly that worth.

BTW:

dewster, the Vintaudio C7 is one of my favorites but when I've compared it (head to head) with our Clavinova CLP-170 (a very old model based on a sampled CFIIIs I believe) I would still rather play the Clavinova; but I sure like recording with the Vintaudio C7.

Oh and I'm not on the payroll or receive endorsements from any DP or sample vendor (although I have beta tested a few).

Rodney

You mean Yamaha would actually pay someone to demean and dissuade Dewster?

I don't believe a word of it.

Snazzy

By the way, Jim, unlike Dewster, Galileo was a pro.

Okey dokey Sally.

Ok, so it's Sally instead of Jim...no problem.

Thanks Sally.

Snazzy

Thanks Rodney! Excellent!

And thanks Dewster for the solid foundation with the numbers and the graphs!

Very well said, Rodney...I feel the same way about VST's and acoustic grands, although I play strictly digital pianos now.

Snazzy

Excellent results from the HP-307. I assume a similar result would be obtained from the new Supernatural upgrade kit for the RD-700GX...would make a powerful case for itself against much of the competition...assuming of course that it was pleasing subjectively to enough people.

Thanks Dewster,

Steve

There is a new DPBSD MIDI file now up at the share point:

====================
= Revision History =
====================
v1.4 - 2010-02-17:
- Added C5 (x7) velocity = 127 at start of file as an aid in setting peak recording level.
- Added two pedal=111 notes to the partial pedaling test, moved all pedal events midway between the notes.


I also updated the readme file to include the following new recording instruction:

Recording & MP3 output:
- If you record directly to MP3, please don't normalize the file or do any other processing to it.

Rodney, thank you for your excellent post, and helping to restore a degree of order to the proceedings.

Great job!

Cheers,
James
x

What would stop DP manufacturers from keep using whatever hardware they're using now, but simply adding a small SSD and streaming the samples from the SSD? SSDs are light, noise free and durable (as long as you mostly read and don't write so much) and so should meet all requirements. Random read speed is about 50MB/s. Shouldn't that be plenty fast enough? And the end user price of an Intel 40GB SSD is just 120 USD.

Whaa? Sorry, no FUD emanating from my camp. Though one look at the sheer mountains of FUD around here and I can see how someone might make the logical assumption that there's no way just one camp could produce it all. But there you go.

Oh, yes there is.

Snazzy

Nothing's changed for me I'm still watching the thread keenly with a wry amusement.

Anything that tries to objectify a subjective always makes me PMSL.

I knew it would be worth to wait - I wanted to buy a new DP since july last year but decided to wait for new Roland models

Im almost sure I turned reverb off but I guess I could do another recording to be sure.

The piano has settings like:
- damper resonance
- duplex scale
- string resonance
- key off resonance
- damper noise
They are all set to the default middle (5/10) and the effects are audible. If you'd like I can change something for testing, but maybe just my input levels are too low? See here original file:
http://www.mediafire.com/?yvnmqzjnznj

Do you know how to adjust it in Adobe Audition (only trial but fully functional I think)? The slider in windows mixer is set to highest level, on my piano the volume was almost maxed. My audio card is Asus Xonar DX, connected from piano's out ports into line in.

Roland U.S. now has the HP-307 manual online at:

http://media.rolandus.com/manuals/HP-307_OM.pdf

Page 25 talks about "Reverb" and it has a setting from 1-10.

I hope to have my 307 in another 2 weeks. So I've already browsed through the manual.

Sally will have something negative to say about this, I am sure.

Also, dewster mentioned that it didn't sound like reverb was turned off -- I came across this paragraph in the HP-307 manual:



Sally will have something negative to say about this, I am sure.

Congratulations on your choice of the HP-307, Sally...I hear it is a great instrument.

Snazzy