I'm surprised DFD streaming isn't more popular with ROMplers

[Sundown]

 

...especially with the advent of Solid State drives.

 

The only board I can think of that might support disk streaming is the Kronos (and I'm not even sure of that). Given the acoustic realism gap between a software sampler on a PC/Mac versus a typical ROMpler, I'm surprised more manufacturers haven't adopted it. The amount of sample RAM that a soft sampler can stack up these days (with different articulations, etc.) is stunning.

 

So what's the bottleneck? Is it the slight wait time between changing patches? Is it a durability concern?

 

Just curious --

 

[AnotherScott]

I think it may be that streaming is a non-trivial thing to do. AFAIK, everything that streams is built either on Linux, Windows, or OS X. So unless you're building something on one of those platforms (which involves its own complications), you'd have to do some pretty deep OS programming of your own.

[J. Dan]

My Akai 3000XL sampler did 2 tracks of streaming from a SCSI connected ext HD well over a decade ago.

[jerrythek]

My Akai 3000XL sampler did 2 tracks of streaming from a SCSI connected ext HD well over a decade ago.

 

As did the Ensoniq ASR-10 V2, released in early 1996 (not quite sure of the date).

 

Regards,

 

Jerry

[J. Dan]

My Akai 3000XL sampler did 2 tracks of streaming from a SCSI connected ext HD well over a decade ago.

 

As did the Ensoniq ASR-10 V2, released in early 1996 (not quite sure of the date).

 

Regards,

 

Jerry

 

Jerry, just FYI, I'm "LZ" with the Class X logo over on Korg Forums. You helped me understand a lot on the Kronos. I really think if they could let go of the "old way", the new way could be done in a much more intuitive way. It's a matter of balancing backward compatibility with new features - I get it....but it sure does confuse a lot of people. That said, things are moving in the right direction, and streaming on the Kronos is huge.

 

 

edit: as well, of course, as BurningBusch and DanAtKorg

[jerrythek]

My Akai 3000XL sampler did 2 tracks of streaming from a SCSI connected ext HD well over a decade ago.

 

As did the Ensoniq ASR-10 V2, released in early 1996 (not quite sure of the date).

 

Regards,

 

Jerry

 

Jerry, just FYI, I'm "LZ" with the Class X logo over on Korg Forums. You helped me understand a lot on the Kronos. I really think if they could let go of the "old way", the new way could be done in a much more intuitive way. It's a matter of balancing backward compatibility with new features - I get it....but it sure does confuse a lot of people. That said, things are moving in the right direction, and streaming on the Kronos is huge.

 

Glad to help you out - wherever. Not sure what old way/new way you're talking about... related to streaming, or architecture in general? We shouldn't take this thread too far off topic...

 

Jerry

[J. Dan]

Old was not streaming. When they added streaming, they had to retain the old file structure/preload to be backward compatible. So you now kind of have 2 ways - preloading or using the _userbank file to stream, yet you have to load to edit....then the different methods of pointing programs to multisamples and samples. Old way is preload samples. New way is point to _userbank and stream.

[mate stubb]

I had an Alesis Fusion once. I thought it had streaming from disk but I might be wrong.

[J. Dan]

I had an Alesis Fusion once. I thought it had streaming from disk but I might be wrong.

 

Didn't stream, but it handled memory FIFO with no preload. Select a patch, it loaded those samples into memory - no preload. Didn't stream, just on demand.

[AnotherScott]

My Akai 3000XL sampler did 2 tracks of streaming from a SCSI connected ext HD well over a decade ago.

That's not the same as what I (or, I think, the OP) was talking about.

 

There is a much simpler kind of "streaming" where you're just talking about a single mono or stereo sound file. (The original iPod did that, too, inexpensively, also over a decade ago.)

 

But what a Kronos (or a PC) can do is different, being able to stream, for example, any of 88 stereo files within milliseconds of being triggered, while continuing to play all the previously invoked streams as needed, with no glitches. Maybe it's a matter of sheer horsepower, or maybe, as I suggested, a difficult OS task.

 

Kurzweil has apparently patented a new approach to doing this in the Forte. And I guess Roland is doing this, to a limited extent, with the sample pads in the FA. (I understand the FA can trigger long samples off the flash, I don't know how many such samples it can stream simultaneously... all 16? Not that I expect that there's a lot of musical need for that...)

[xp50player]

88 stereo files? Every sample in the keyboard (12Gb) is available instantly, because it preloads them all on startup. Otherwise, there would be a delay when changing Programs/Combis.

 

the Bottleneck, for the OP, as far as the Kronos, is the boot time for the embedded Linux + the preload time for the factory sounds, + any sample libraries you have added. With the Kurzweil Forte, they are able to do everything with DSP chips, and store the preload in ROM for the massive streaming piano samples. But it is a closed system, no sample expansion. Roland has made the sample ROM size spec obsolete by their proprietary resynthesized SuperNatural sounds, which are continuously variable throughout the velocity range and across the keyboard.

[JeffLearman]

88 stereo files? Every sample in the keyboard (12Gb) is available instantly, because it preloads them all on startup. Otherwise, there would be a delay when changing Programs/Combis.

That is the opposite of "streaming from disk." If the Kronos supports streaming, then it doesn't have to preload full samples for any patch that uses DFD (direct-from-disk) samples.

 

To handle DFD, a player has to preload the first [x] amount of samples, where [x] is however many samples to cover disk access time to get to the remaining data. So, for a very oversimplified example, if your disk drive has a 10 ms maximum seek time, you'd want about 10 ms of every sample preloaded in memory, to support DFD streaming. But only 10ms of each sample.

 

Nemesys (GigaSampler/GigaStudio, sold to Tascam) had a patent on this, but I don't know if they've defended it.

 

BTW, it could be way more than 88 samples. For example, Scarbee CEP uses 16 velocity layers per note, so 16 * 88 (mono) samples.

[AnotherScott]

88 stereo files? Every sample in the keyboard (12Gb) is available instantly, because it preloads them all on startup. Otherwise, there would be a delay when changing Programs/Combis.

As Jeff said, only the initial bits of the samples are preloaded on startup, not the entire sample. To load 12 GB of samples at startup, there would have to be 12 GB of available RAM (and then, of course, streaming from disk is no longer necessary). That's the point of streaming from disk... only a small portion is preloaded into RAM, the rest is called upon to be streamed in real time from disk as needed, so large libraries of samples can be played without comparably large amounts of RAM. That's the function that, prior to the Forte, seems only to have been available under Linux, Windows, or OS X. (Assuming that Forte is not Linux based.)

[Sundown]

The streaming I'm referring to is Gigasampler-style (or HALion, Kontakt, Mach Five, etc).

 

A small amount of sample data is loaded to RAM, and the rest is streamed off of a hard drive in real time. It allows for *very* large sample sets.

 

HALion for me loads quite quickly, and the realism is incredible.

[burningbusch]

Roland has made the sample ROM size spec obsolete by their proprietary resynthesized SuperNatural sounds, which are continuously variable throughout the velocity range and across the keyboard.

 

Wish that it were true. I have the Integra-7 which has the largest number of SN sounds in any of the Roland products. While SN does a good job overall with timbre transition, mellow to brighter, there are other aspects of the sounds that are not as successful. It's pretty obvious to my ear that all the source sounds are sample-based. On many sounds, you can hear the static loop come in very shortly after the initial attack. This is most obvious on the clavs, ac. guitars, bells, mallets. Also, samples are stretched across multiple notes--listen to the EPs, ac. guitars, etc. The decay envelopes are artificial and linear. I think the winds, brass, strings, elec. guitars come off quite well and for the most part are not affected by the above issues. But it does tick me off, as an owner, the Roland didn't go the extra mile on some of these sounds that would have clearly benefitted from better or more complete sample material. In the end there aren't a huge number of SN acoustic tones and why they couldn't have made each one the best it could be is frustrating.

 

So no, in my opinion, Roland has not made detailed sampling obsolete. The best, most detailed, most realistic acoustic sounds I have still come from uncompromised samples.

 

Busch.

[burningbusch]

So what's the bottleneck? Is it the slight wait time between changing patches? Is it a durability concern?

 

Just curious --

 

The Kronos is lightning quick when changing programs. Durability is not an issue. As AnotherScott pointed out it's not trivial to pull this off. The biggest downside is the 2+ minute boot time which is part OS and part pre-load of the samples.

 

So yes only the Kronos and now the Kurzweil Forte allow for streaming of samples.

 

Busch.