195,058 users have contributed to 42,961 threads and 258,119 posts.
In the past 24 hours, we have 8 new thread(s), 38 new post(s) and 75 new user(s).
Welcome llGoDLikEll,
Vienna Instruments are recorded in 96 kHz with 24 bit resolution. All recordings are stored and processed in 32 bit FP resolution. In a final step, the samples are converted to 44.1 kHz / 24 bit just before mapping, using a proprietary SRC tool. To reduce their size, the data of intruments is compressed (not reduced) before delivery.
HTH,
Thank you very much for the detailed description.
But it will be awesome if there is a 96k/24bit version for more precise work
[color=#000000]Welcome[url=/community/profile/122673-llGoDLikEll]llGoDLikEll[/url],[/color] [color=#000000]Vienna Instruments are recorded in 96 kHz with 24 bit resolution. All recordingsare stored and processed in 32 bit FP resolution. In a final step, the samples are converted to 44.1 kHz / 24 bit just before mapping, using a proprietary SRC tool. To reduce their size, the data of intruments is compressed (not reduced) before delivery.[/color] [color=#000000]HTH,[/color]Thank you very much for thedetailed description. But itwill be awesome if there is a 96k/24bit version for more precisework I think so, but it will be a huge file😊 48k may be more realistic. Of course, if there is 96k I will happy to prepare 1TB SSD
96 Khz @ 24 bit would be quite nice, and would bring lower latencies, which can be significant for playback 'feel'. Latencies are cumulative and live recording sampling rate is determined by the lowest rate of any part of the chain. So the net effect of VSL sticking to a 1980's era sampling rate is a doubling of the latencies of every part of the playback chain.
A 2 TB NVME PCIe SSD is $200. 32 GB of DDR4 RAM for a laptop is $150.
Not sure why VSL has decided to choke the sound chain latency by sticking to 44/48 Khz.
Actually, using 24/96 samples will slow down the system. Without adding any perceivable quality.
Paolo
There are 2 different ways to consider 'slowness'.
1. How slow is the computer hardware doing the processing, which is how you seem to consider this issue. Of course, if one doubles the sample rate then processing power and bandwidth must also be double or the system will 'slow down'. However, almost all laptops or desktops of a mid-grade specification (~$1000), and are less than 5 years old, already have excessive amounts of resources and will not be affected by 96 or even 192/384 Khz sampling rate. Our hardware is fast enough to not be slowed by increasing sample frequency.
2. What is affected by the other type of 'slowness' is the piano player having to deal with the lagtime or delay that occurs in the time period from pressing a key to hearing a sound; this is referred to as latency. And latency absolutely affects the quality of the playing experience. The lower the sampling rate the higher the latency, and the higher the latency/delay the more the player has to adjust to the different 'feel' and 'sound'.
Certain players feel the affects of latency at 3 ms, and there have been many studies into psychoacoustics which support the idea that such low latency is perceived and affects musicians. A 44 Khz sampling rate makes it more difficult to achieve good sound quality and low lagtime/latency in the instrument.
Given thier ownership of awesome acoustical grands, robotic sampling system for the same, access to a world class studio, and having ample technical resources, VSL is in an excellent position to release an engineering study on latency and include it in their product information and customer education materials. It would be great to see data on the internal latency of the Synchron player, and then see how that sampling rate interacts with other parts of the processing chain and thereby affects total system latency.
If VSL doesn't do it...maybe I'll do it in the living room? Or can I get an invite to Vienna? 😉
On a concluding note, if one really wants to look at the necessity or value of the VSL library configuration, I would suggest that 24 bits of dynamic range for an acoustical instrument is ABSOLUTELY overkill when considering most scenarios. 24 bits equals 144 dB of dynamic range, and I dare say that even the most enthusiastic pianist can hardly play the Steinway 274 at a level that competes with jet engines at 100 feet or machine gun fire at 1 meter. Of course, there is the convenience of laying down 24 bit tracks that coexist with live mics that do not have their gain properly adjusted, but that is convenience only, while 96+ Khz sample rate potentially has a much more fundamental impact on playability. And of course, 24 bits does give MIXING headroom, but that is not required for SAMPLING of a grand piano, so VSL gave themselves much more headroom than could possibly be useful. Even 16 bits is excessive dynamic range for a piano library.
BTW, Pianoteq Pro includes 96 Khz sampling rate and also includes the Steinway D in their library. Could make for an interesting comparison between sampled and modeled sound. FWIW, I think the VSL would come out ahead at the same sampling frequency, but would not prevail for live playability if having double the latency.
Given that VSL is recording at 96 Khz (with 32bits!) it really is amazing that there is no 96 Khz release that exists alongside the 44/48 Khz release. Can we get an explanation for that decision and possibly some analysis of VSL's internal latency?
But you are not forced to use 24/96 samples to set your audio interface to 24/96. You can still play lower resolution/frequency samples, and still enjoy the lower latency warranted by higher system settings.
I agree that nowadays computers are very powerful, but when dealing with all the instruments of an orchestra, maybe with multiple microphones, and additional processing, you'll end up with any system chocking. Going the for a an excellent 24/44.1 would give CD quality, while keeping the system relatively light.
Paolo
But you are not forced to use 24/96 samples to set your audio interface to 24/96. You can still play lower resolution/frequency samples, and still enjoy the lower latency warranted by higher system settings.
Paolo
I agree that there are scenarios where higher sampling rate would impose excess processing or storage. But the approach of increasing sampling rate after the VSL library actually increases latency, as upsampling the frequency rate requires another processing step.
Better that VSL gives us the option of a higher sample rate library. Then, even in scenarios requiring lower sample rates for recording the orchestra, etc, we could just use a few processor cycles to downsample the frequency rate without even increasing the latency (this only works for downsampling).
FWIW, I would like to see some double blind latency tests with varying freqency rate. I would not be surprised at all if we have 'super hearers' that can process sound much faster than average people...and my guess is that some very gifted musicians might actually be able to process better with lower levels of latency.
I am trying to systematically eliminate the negative effects on sound quailty and differences in playability between 'mechanically voiced' and 'electronically voiced' grand pianos. VSL has potentially done much to eliminate potential negative effects, and the 96 Khz might just be another step towards bridging the divide.
Just my opinion, and I've been known to speak with my foot in my mouth many times, but this is my understanding (I earned my B.M. degree in Music Production and Engineering years ago when digital audio was new.)
You’re saying that systems these days are *optimized* for sample rates like 96KhZ @ 24-bits… to the point where input, processing, and output at that rate is actually *faster* than it would be if it had to digitally convert half as much data? Hmmm. You mention a few clock cycles to convert. I’m not a CPU expert nor an Apogee employee, but my understanding is that latency in one's system depends on I/O, processing, and conversion (at least). If I were a betting man I’d guess the nanoseconds it’d take interpolating half as much data would ballpark-equivalent to natively-processing twice as much data.
Just my opinion, and I've been known to speak with my foot in my mouth many times, but this is my understanding (I earned my B.M. degree in Music Production and Engineering years ago when digital audio was new.)
You’re saying that systems these days are *optimized* for sample rates like 96KhZ @ 24-bits… to the point where input, processing, and output at that rate is actually *faster* than it would be if it had to digitally convert half as much data? Hmmm. You mention a few clock cycles to convert. I’m not a CPU expert nor an Apogee employee, but my understanding is that latency in one's system depends on I/O, processing, and conversion (at least). If I were a betting man I’d guess the nanoseconds it’d take interpolating half as much data would ballpark-equivalent to natively-processing twice as much data.
Not to mention the memory footprint. A full orchestral piece requires 60 plus gigabytes of ram for me, and that is really nothing in comparison with what it could be. There is a reason why 44.1 at a 16 bit depth rate was chosen for CD audio, this is considered just above the maximum quality that anyone can hear a difference. Anything above this is really unecessary, unless you are time stretching samples (think slow motion video, the more frames you have, the smoother it will be). When it comes to orchestral, I dont think time stretching samples is something many people are apt to do.
Not to mention the memory footprint. A full orchestral piece requires 60 plus gigabytes of ram for me, and that is really nothing in comparison with what it could be. There is a reason why 44.1 at a 16 bit depth rate was chosen for CD audio, this is considered just above the maximum quality that anyone can hear a difference. Anything above this is really unecessary, unless you are time stretching samples (think slow motion video, the more frames you have, the smoother it will be). When it comes to orchestral, I dont think time stretching samples is something many people are apt to do.
1. No disrespect, but have you read the thread? This is about LIVE instrument playability quality, not RECORDED/MIDI music file playback sound quality or time stretching.
To summarize, there is a negative effect of latency on pianist's sense of hearing and motor cortex processing as impacted by low sampling frequency. That is, if you press a key and you expect to hear the sound in 3 Milliseconds, but the sound instead comes to you in 6 Milliseconds, that will be as if the sound board has moved from 3 feet away to 6 feet away. A bit wierd to say the least. And that assumes no other latency elsewhere in the sound processing chain. As there always is this latency in every component, the cumulative effect of latency might mean that the piano sounds as if the sound board is 10 feet or more away. And that's before we start to consider the effect of jitter, which can be described as the sound board constantly shifting back and forth several feet, with this perceived movement happening from note to note or even during the resonance of a single note. Aye...very bad indeed for your brain to be 'hearing' the grand piano vibrating back and forth.
Also note that as this jitter is likely random, good luck trying to achieve consistent playback between pieces. And that is true even for playback of MIDI notes in which instruments are closely placed.
There is a way to elilminate the 'sound shifting': By roughly doubling the frequency from 44 to 96 Khz, the latency of the VSL library is cut in half, and that allows for a more consistent and authentic playing experience. And by doubling again from 96 Khz to 192 Khz the latency is halved again and we have more 'latency headroom' to stay under the audibility threshold.
Don't take my word for it, but do take VSL's deafening silience on this topic as being problematic. I suggest if my explanations are not sufficient, please research and understand other trusted sources.
2. As to RAM requirements, this piano at 24 bit 44.1 Khz currently has a RAM requirement of 4-6 GB. Doubling the sampling frequency will half the latency and only require a total of 6 GB * (96/44.1) = 13 GB. Most laptops with any sort of multimedia claims will have 16 GB RAM or more.
To editorialize a bit, VSL is doing an enormous dis-service to themselves and their customers by attempting to ignore the issue. Unfortunately, an uneducated customer at some point becomes a frustrated customer...and that's a net negative for musical creativity and classical music as a whole (LONG LIVE POP MUSIC?).
[...] By roughly doubling the frequency from 44 to 96 Khz, the latency of the VSL library is cut in half, and that allows for a more consistent and authentic playing experience. And by doubling again from 96 Khz to 192 Khz the latency is halved again and we have more 'latency headroom' to stay under the audibility threshold. [...]
Sorry to ask the obvious: Why don't you change your audio system's driver to those preferred sample rates if you want to bring down the latency?
Hi all,
About space, I agree that it is no more a problem today (including M2).
About latency, I need to study it deeper…
But let’s talk about sound quality: I did many tests about 44.1 vs 96 kHz with synth sources, and I clearly make a difference (not really with 192kHz, I’m not totally sure).
But it was (reproducible) synth sounds.
I’m not engineer so, three questions here about samples :
Does a mic source will make a perceptible difference between 96 and 44.1kHz ?
Is 96kHz better than 44.1kHz for mixing voices ? (Considering that a sample player is also a mixer.)
And what about processing ? Is is better to process an effect in 96 than 44.1kHz ?
If yes, a 96kHz library would be welcome anyway.
Regards
Gabriel Plalame
