Jump to content


Please note: You can easily log in to MPN using your Facebook account!

Rates of improvement?


Tusker

Recommended Posts

Originally posted by Is There Gas in the Car?:

We are all spoiled - ain't technology great?

By the way, your avatar doesn't show up anymore since days. :rolleyes: Maybe it's time to fix the problem? ;):P

 

Have a glass of Absinthe for me. :)

Link to comment
Share on other sites



  • Replies 52
  • Created
  • Last Reply
Originally posted by Cydonia:

By the way, your avatar doesn't show up anymore since days. :rolleyes: Maybe it's time to fix the problem? ;):P

 

Have a glass of Absinthe for me. :)

The problem is on your end, I believe. It's working fine for me.

 

You got something against sheep? :(;)

"Music expresses that which cannot be put into words and that which cannot remain silent." - Victor Hugo
Link to comment
Share on other sites

Originally posted by Tusker:

[When it comes to perception, I also think there is a "bench test" mentality that is similar to computers. You can run mandelbrot equations to test computers, but most of us want to do wordprocessing and play some games.

My take on this is that at least some subjective perceptions of sound are based on qualities that can be objectively measured. "Pure," "warm," "rich," "full," "brittle," "harsh," "thin," etc. - I think most of these boil down to measurable, definable issues such as aliasing, jitter (oscillators can jitter too!), saturation/distortion, frequency response, etc. Just my opinion, though.

 

Originally posted by Tusker:Similarly for aliasing tests ... all digital synths alias at some point. If you really want a saw wave to sound clear 4 octaves about middle c, just use analog and you are done.
They'll all alias eventually (at least with current technologies), but - as shown by the tests referenced elsewhere in this thread - some will alias much, much more than others. The effects may be subtle and limited to higher pitches, as you suggest - or they may take over the sound entirely in more moderate pitch ranges, as happened to me fairly recently when I enabled oscillator sync on a popular digital synth. Aliasing may also become more evident when sounds are processed through distortion/overdrive (on top of the aliasing added by the distortion). So, it's not *necessarily* a subtle issue.

 

- Dan

Dan Phillips

Manager of Product Development, Korg R&D

Link to comment
Share on other sites

Originally posted by Is There Gas in the Car?:

The problem is on your end, I believe. It's working fine for me.

I remember your sheep avatar, but now I see a little red X. Hmmmm... :confused:

 

Maybe I drank too much Absinthe. :D

Link to comment
Share on other sites

Originally posted by Cydonia:

Originally posted by Is There Gas in the Car?:

The problem is on your end, I believe. It's working fine for me.

I remember your sheep avatar, but now I see a little red X. Hmmmm... :confused:

 

Maybe I drank too much Absinthe. :D

In the spirit of objective testing, I'm now sipping a bit of Absinthe ( L\'Esprit du Vieux Carre ). I can confirm that I do indeed see the sheep. I think that's a good thing. :-)

 

- Dan

Dan Phillips

Manager of Product Development, Korg R&D

Link to comment
Share on other sites

Originally posted by Dan Phillips:

My take on this is that at least some subjective perceptions of sound are based on qualities that can be objectively measured. "Pure," "warm," "rich," "full," "brittle," "harsh," "thin," etc. - I think most of these boil down to measurable, definable issues such as aliasing, jitter (oscillators can jitter too!), saturation/distortion, frequency response, etc. Just my opinion, though.

I agree--the trouble is that for many it's hard just to understand what the terms mean, much less how they relate to perceptual characteristics. It's not helped that terms such as distortion make sense for certain applications (e.g., an amplifier ideally is a "straight wire with gain"*) but with a musical instrument? If we didn't like "distortion," we'd listen to sine waves; the very point of an instrument is to "colour" the sound in pleasant ways producing its characteristic timbre.

 

On the other hand, something like aliasing sounds bad--unless you specifically want it for an effect or grit.

 

I wonder how you "measure" the sound of a filter? Why do some filters or filter settings, for example, sound "squlechy"?

Link to comment
Share on other sites

Originally posted by soundscape:

I wonder how you "measure" the sound of a filter? Why do some filters or filter settings, for example, sound "squlechy"?

If "squelchy" means "wet", here's Jim Clark's answer:

 

In my studies of sonic alchemy I have come to the conclusion that the wetness of wet filters comes from 'dispersion' which is a technical term for different frequencies having different delays. (the term comes from the effect of a prism on white light in dispersing the lights spectrum). Adding in frequency dependent delays definitely makes sounds 'wet'. Typically delays up to 100 msec or so sound good and wet, but the Nord Modular can only do delays up to about 10msec. Nonetheless, this is enough to moisten the sound somewhat. Check out the patch shown below, which dampens the sound of a percussion module. Twisting knob 2 takes the sound from 'dry' to 'wet'.

 

Link here:

 

http://www.cim.mcgill.ca/~clark/nordmodularbook/nm_filters.html

Link to comment
Share on other sites

Dan, thanks for taking the time for your comments. I learn a lot from conversations like these.

 

Originally posted by Dan Phillips:

My take on this is that at least some subjective perceptions of sound are based on qualities that can be objectively measured. "Pure," "warm," "rich," "full," "brittle," "harsh," "thin," etc. - I think most of these boil down to measurable, definable issues such as aliasing, jitter (oscillators can jitter too!), saturation/distortion, frequency response, etc. Just my opinion, though.

Well it's a very well informed opinion. To try to get our conversation past a half-empty versus half-full discussion, let me ask you how complete the cause and effect DSP model is? For example we have been studying analog timbres right? We know that digital problems such as jitter, aliasing etc, can detract from an analog model. We also know that "analog impurities" need to be added back in an elegant way (I love the pre-filter distortion on the OASYS model btw. It enables a whole class of desirable sounds very simply, in a way that post-filter distortion doesn't.)

 

But we are never fully conscious of the impact of our design choices are we? I would suggest that any time we make more than a thousand design decisions there will always be combined effects that we could not predict. These effects don't live in the area of "problems" but more in the area of "character." I see our peers talking about sonic character and I know that each manufacturer has a something of a signature. It's not good or bad, just different.

 

Recently we had people talking about how warm a (4 MEG?) Kurzweil piano from the 1990s sounded. Others talking about how EMU samplers make samples sound fuller. Sure it has to do with frequency response and quantifiable factors, but I don't see manufacturers adjusting their timbral signature. Some might say it's because they don't want to. I'd say it's part of their subconsious sonic signature. The product of several decisions acting systemically.

 

It's in that realm (not across the board) that the benchmarks fall down for me. Not because the benchmarks aren't relevant or useful. They certainly are and as our cause and effect model becomes more complete they will be more useful. But they don't tell the whole story. Because the cause and effect model of DSP is not complete yet.

 

I'd say the day we hear a well modelled symphony orchestra, the model would be close to complete. We would have modelled several classes of excitors, resonators, their second order interactions and the environments in which human ears experience them.

 

Originally posted by Tusker:Similarly for aliasing tests ... all digital synths alias at some point. If you really want a saw wave to sound clear 4 octaves about middle c, just use analog and you are done.
[/QB]

 

Dan I agree with you about the difference in the design of dsp algorithms. I didn't mean to say otherwise. Sorry if I misspoke. Aliasing is certainly an important nut to crack from a manufacturer's perspective. I have wacko sounds from stressing some of the dsps on my trusty Roland rompler. (see I can mention names. ;) )

 

I was speaking from a player's perspective ... I was just trying to say that we should use the best tool for a particular job. Each tool has it's sweet spot. It's silly to wait for an instrument with an infinite sweet spot. The tools we have been given do so much more than previous generations could have imagined ...

 

Jerry

Link to comment
Share on other sites

Originally posted by Tusker:

There's some interesting information on that thread. Thanks. I too have been turning to synthesis in lieu of larger samples. Elhardt does a lot of cool stuff in this area. Here is his synthesized piano done on a Arturia Modular:

 

http://home.att.net/~synth6/Arturia_MMV_Piano.mp3

Wow. That's some impressive piece of programming...!

 

Originally posted by Tusker:

I find these kinds of techniques to be part of the improvement cycle. It's not just the manufacturers, the hardware and the software. It's us as players being ingenious in our use of synth resources. While I don't do emulations on the Nord (I'm no Elhardt) I have enjoyed getting controllable gutty and reedy qualities similar to cello and large bore woodwinds ... just by tuning the filters. Things you don't get from a bigger sample.

This is just it... trouble is that some of the recent patches I've seen use larger samples but don't really take advantage of the synth engine... I really think at least using some synthesis, not just sample playback, that gives the best results.
Link to comment
Share on other sites

Eventhough I own and have worked with GigaStudio, Kontakt, EXSII 24, MachFive, and more I wondered why I never heard the aliasing shown in the test. Then it occurred to me, I don't run 15KHz samples up and down the keyboard. Somehow, I never found a musical application for that. The aliasing test above is a red herring.

 

On the other hand I do use high quality sample libraries from VSL, East West, Scarbee, Synthogy, Native Instruments, Sonic Implants and others. When compared against the compromised samples found on ROMplers, it should be clear to anyone with ears and lacking an agenda that there is a significant quality difference.

 

Years ago, someone posted a very interesting experiment that has stuck with me. A short example taken from a string quartet was used as the source. Two different manipulations were done. In the first case the 16-bit file was edited with a reduce in volume of 3db. The file was saved at 16bit resolution, then edited again with a 3 db increase in volume and again saved at 16-bits. The second file was again a 16-bit source which was lowered 3db in volume but saved as a 24-bit file and again brought back up 3db, saved as 24-bit and finally dithered and saved as 16-bit. The source, 16-bit edited and 24-bit edited were all available in uncompressed 44.1KHz 16-bit format. It was obvious to me that the 24-bit sounded much closer to the original and the 16-bit edited was more strident. It demonstrated to me how delicated digital audio really is. Lowering and rasing 3db is nothing compared to the manipulations that take place as the sample is triggered, filtered, placed in an envelope, EQed, and FXs applied. You cannot manipulate digital audio at the same bit rate without f#@$ing it up. Bob Katz comes to similar conclusions in writings on his website and in his book.

 

When Roland came out with the XV series I still had a JV-1080 available for comparison. Using the JV ROM cards as source material, it became immediately clear that the new XV had clearer highs, more defined lows and better transients using the exact same programs and waveforms. Most people attributed this improvement to the DACs but even when I took the digital out of the XV into a variety of DACs, they all sounded better than the JV. That suggested to me that Roland had cleaned up the internal workings and pathways with the XV series--maybe got rid of some of that 16-bit stuff. Early digital hardware was obviously 16-bit (at best) throughout. At some point this changed, or did it? Who knows. It's all very murky.

 

All this has lead me to have a healthy skepticism with regards to digital hardware (digital synth/samplers and hard disk recorders). I don't trust that they've dotted all the (i)s and crossed all the (t)s. The OASYS would be the exception and possibly the Kurzweil K2500/K2600. The OASYS is computer-based so it's calculations are 32-bit and I believe the Kurzweil uses an open 24-bit DSP architecture. Interesting that most people consider these two synths to have the most pristine output of any hardware synths. Remember how god awful some early digital recorders sounded, only slightly better with outboard DACs.

 

Software is inherently 32-bit so it can work with 16-bit and 24-bit audio without compromise.

 

Busch.

Link to comment
Share on other sites

Originally posted by soundscape:

This is just it... trouble is that some of the recent patches I've seen use larger samples but don't really take advantage of the synth engine... I really think at least using some synthesis, not just sample playback, that gives the best results.

I don't know what this means. What are the "synth engine" aspects of the Yamaha, Roland and Korg ROMplers?

 

Busch.

Link to comment
Share on other sites

Originally posted by burningbusch:

The OASYS is computer-based so it's calculations are 32-bit and I believe the Kurzweil uses an open 24-bit DSP architecture. Interesting that most people consider these two synths to have the most pristine output of any hardware synths.

That illustrates Dan's point that there are objective criteria supporting subjective sonic assessments. Thanks for sharing those examples Busch.

 

Cheers,

 

Jerry

Link to comment
Share on other sites

Originally posted by burningbusch:

Originally posted by soundscape:

This is just it... trouble is that some of the recent patches I've seen use larger samples but don't really take advantage of the synth engine... I really think at least using some synthesis, not just sample playback, that gives the best results.

I don't know what this means. What are the "synth engine" aspects of the Yamaha, Roland and Korg ROMplers?

 

Busch.

PSR-550:

 

http://www.synthmania.com/Audio%20Files/Keyboards/Yamaha/PSR-550/XG%20sounds/615%20Creation.mp3

http://www.synthmania.com/Audio%20Files/Keyboards/Yamaha/PSR-550/XG%20sounds/384%20Acid%20Bass.mp3

Link to comment
Share on other sites

Originally posted by soundscape:

Originally posted by burningbusch:

Originally posted by soundscape:

This is just it... trouble is that some of the recent patches I've seen use larger samples but don't really take advantage of the synth engine... I really think at least using some synthesis, not just sample playback, that gives the best results.

I don't know what this means. What are the "synth engine" aspects of the Yamaha, Roland and Korg ROMplers?

 

Busch.

PSR-550:

 

http://www.synthmania.com/Audio%20Files/Keyboards/Yamaha/PSR-550/XG%20sounds/615%20Creation.mp3

http://www.synthmania.com/Audio%20Files/Keyboards/Yamaha/PSR-550/XG%20sounds/384%20Acid%20Bass.mp3

Samplers have long had filters and envelopes, big deal. ROMplers also have always had them.

 

GigaStudio allows you to apply a specific filter setting, filter type and evelope settings for the left side of a single stereo waveform and a different setting/type/envelope for the right side. This can be done for every sample in the program. Many developers use this to fine tune each sample in the instrument.

 

Have you ever looked at the scripting language in Kontakt 2.0? There is nothing like it, hardware or software. It is being exploited by numerous developers.

 

Busch.

Link to comment
Share on other sites

Well, here's another example:

 

http://www.scarbee-downloads.com/demos/js/jslap_demo1_solo_js1.mp3

 

Impressive stuff, but you can surely hear the flaws of 'concatenative' sampling.

 

Now combine three bass parts and the rest of the track and the results are simply fantastic:

 

http://www.scarbee-downloads.com/demos/js/jslap_demosong_1.mp3

 

 

But then again a bass patch like:

 

http://www.synthmania.com/Synthesizers/Roland/JD-990/Audio/Factory%20Patches%20examples/I-64%20Mo%20Funk%20Bass.mp3

 

Is pretty darned powerful.

 

 

It seems to me worthwhile discuss whether 'concatenative' approaches with different articulations will tend to yield good results (that is, not with flaws that distract) and whether it's worth programming a complex track with articulations.

 

(Naturally the same issues exist on ROMplers, e.g., Yamaha's "MegaVoices.")

Link to comment
Share on other sites

I think what I have to say really isn't about hardware vs. software per se, it is really more about quality, and implementation.

 

For the highest quality audio, everything must be given a high level of scrutiny. Naturally, this applies to hardware as well; I've come across rompler TR-808 drum samples with noise in them for which there is no excuse; some romplers have lacked resonant filters; I've come across multi-samples on romplers where adjacent keys didn't perfectly match and at least one sample had an extraneous sound in it.

 

 

Originally posted by burningbusch:

Years ago, someone posted a very interesting experiment that has stuck with me. A short example taken from a string quartet was used as the source. Two different manipulations were done. In the first case the 16-bit file was edited with a reduce in volume of 3db. The file was saved at 16bit resolution, then edited again with a 3 db increase in volume and again saved at 16-bits. The second file was again a 16-bit source which was lowered 3db in volume but saved as a 24-bit file and again brought back up 3db, saved as 24-bit and finally dithered and saved as 16-bit. The source, 16-bit edited and 24-bit edited were all available in uncompressed 44.1KHz 16-bit format. It was obvious to me that the 24-bit sounded much closer to the original and the 16-bit edited was more strident. It demonstrated to me how delicated digital audio really is. Lowering and rasing 3db is nothing compared to the manipulations that take place as the sample is triggered, filtered, placed in an envelope, EQed, and FXs applied. You cannot manipulate digital audio at the same bit rate without f#@$ing it up. Bob Katz comes to similar conclusions in writings on his website and in his book.

It's well known that ideally audio processing takes place at greater than 16-bit. But, having said that, the artefacts that occur in a synth (or an outboard processor, for that matter) may well be "pushed down" because the level they're mixed at isn't 0dBFS.

 

Dithering is a good example of implementation detail. You could have a 24-bit source and decimate it down to 16-bits without dithering... and still claim it's "CD quality."

 

Originally posted by burningbusch:

When Roland came out with the XV series I still had a JV-1080 available for comparison. Using the JV ROM cards as source material, it became immediately clear that the new XV had clearer highs, more defined lows and better transients using the exact same programs and waveforms. Most people attributed this improvement to the DACs but even when I took the digital out of the XV into a variety of DACs, they all sounded better than the JV. That suggested to me that Roland had cleaned up the internal workings and pathways with the XV series--maybe got rid of some of that 16-bit stuff. Early digital hardware was obviously 16-bit (at best) throughout. At some point this changed, or did it? Who knows. It's all very murky.

There could be all sorts of reasons for the change, one could even speculate that there's a "hidden" harmonic exciter in the signal path...

 

After all, IIRC according to Eric Persing most Roland synths have used 32KHz samples except for the JD800/990 which use 44.1KHz samples.

 

AFAIK some of the early digital hardware used techniques such as 10-bit companded audio. Yikes! Despite this, there are many mid-80s recordings that sound polished and burst right out of the speakers.

 

The trouble is though that "value engineering" decisions may be sensible in terms of tradeoffs... in software they may not "have" to be made as the developer doesn't have to "care" about fixed resources and throwing polyphony down the drain.

As Dan Phillips pointed out in another forum, the Korg OASYS filters are resonant right up to 20KHz even though the sample rate is 48KHz, whereas another synth runs at 96KHz but the filters aren't resonant beyond 16KHz. (http://www.sweetwater.com/forum/archive/index.php/t-8207.html)

 

Besides, having something like 32-bit floating point processing gets you nowhere if there is audible aliasing; surface specs don't necessarily tell the whole story.

 

Originally posted by burningbusch:

All this has lead me to have a healthy skepticism with regards to digital hardware (digital synth/samplers and hard disk recorders).

Such skepticism should apply to ALL audio products... software or hardware. There has always been few first rate instruments and other audio products and lots of junk. This is why marketing claims of "product of the week," whether made by the most respected companies or not, should be ignored and each product carefully evaluated.
Link to comment
Share on other sites

Here's another point... Erik Norlander on the Fairlight 'Vox' sound:

 

"As far as recreating that sound ... whew. That's a whole other discussion. Many have tried. Get a single singer to give you a very breathy "ahh" voice, record it evenly and in tune and boost the living daylights out of the high frequencies, whilst simultaneously compressing it into oblivion. Get the EQ and compression to distort in just the most unique and pleasing way, placing the sound perception-wise somewhere between a voice and a pipe instrument. Yikes. When you call me to do that session, let me first please make sure I'm previously booked for that day. [Wink]"

 

[http://www.musicplayer.com/cgi-bin/ultimatebb.cgi?/ubb/get_topic/f/18/t/018285.html#000016]

Link to comment
Share on other sites

Originally posted by burningbusch:

Eventhough I own and have worked with GigaStudio, Kontakt, EXSII 24, MachFive, and more I wondered why I never heard the aliasing shown in the test. Then it occurred to me, I don't run 15KHz samples up and down the keyboard. Somehow, I never found a musical application for that. The aliasing test above is a red herring.

You may be right, but aliasing performance is going to be an issue in more 'pedestrian' applications, and this could be a sign of sloppy engineering. GigaSampler does get good results though.

 

I would strongly welcome more tests to obtain 'concrete' information; why don't any magazines publish this sort of thing?

Link to comment
Share on other sites

Originally posted by soundscape:

Originally posted by burningbusch:

[qb] Eventhough I own and have worked with GigaStudio, Kontakt, EXSII 24, MachFive, and more I wondered why I never heard the aliasing shown in the test. Then it occurred to me, I don't run 15KHz samples up and down the keyboard. Somehow, I never found a musical application for that. The aliasing test above is a red herring.

You may be right, but aliasing performance is going to be an issue in more 'pedestrian' applications,
Exactly. For instance, the fundamental pitch doesn't need to be high in order to get aliasing; if it's a really bright sound, the frequency content may stretch right up to the nyquist, regardless of the fundamental. Any pitch change then opens you up to aliasing - and that includes small tuning corrections and playing back at a different sample rate (for instance, playing a sample recorded at 31.25kHz in a 44.1kHz or 48kHz system), in addition to transposition, pitch modulation, etc.

 

Originally posted by soundscape:

I would strongly welcome more tests to obtain 'concrete' information; why don't any magazines publish this sort of thing?

That's something I'd love to see.

 

- Dan

Dan Phillips

Manager of Product Development, Korg R&D

Link to comment
Share on other sites

Originally posted by cnegrad:

I specified that when comparing _purely sampled sounds in romplers_ (without any synthesis involved) to those in large sample libraries, the large sample libs will come out on top. And as a result, the _purely sampled sounds_ in a rompler will make the rompler sound poor by comparison. In my original post I went out of my way to exclude synthesis from my comments.

Isn't it missing the point to use a synth for 'purely sampled sounds' though...? And what about something like the GEM FADE filters, how do you assess the 'purely sampled sound'?

 

I certainly agree that a rompler may not sound the same as a mega sample library with multiple layers, lack of looping, multiple articulations, etc., although some may try and Yamaha for instance have tried the latter with their 'MegaVoices.' The trouble is whether one really sounds 'better' than the other... the more samples, the more trouble there is matching and perfecting them, and even properly 'matched' multiple articulations may still end up with an obviously 'concatenated' sound. Regardless of hardware vs. software, I think it is worthwhile to consider whether quality and expressiveness through synthesis or multiple samples gives better results.

 

 

Originally posted by cnegrad:

I'm also not thrilled about the fact that we've grown accostomed to the new mindset of, "well it's good enough for live use". If I'm going to spend thousands on a hardware board, I want it's sampled sounds to be good enough for the studio.

I suspect the current generation of rompler is stuck in a sort of no-man's land; not enough ROM to 'compete' with the mega libraries yet a loss of the 'hobbyist' market to soft-synths puts synthesis to one side. (We can see this start to change with Korg OASYS.) Nevertheless for out of the box 'hit' sounds something like a Korg Triton is still great, which is why it is used by the likes of The Neptunes (I don't like their music, but they are successful.) Then again Pharrell Williams is still using a Korg 01/W for sequencing!
Link to comment
Share on other sites

Originally posted by cnegrad:

Originally posted by Tusker:

It used to be that way. There's a bunch of 1990s CD's with rompler pianos, string beds etc. Even the Korg M1 piano is out there on a thousand CD's.

But back then those sounds were all that was available, unless you were one of the lucky ones who had access to a Fairlight or Synclavier. Now, original recordings use the high-end libraries, and the romplers just can't keep up, unless you're using a Neko or a Receptor. It's tough now to sit down at my gig at my Roland rompler after playing giga sized high end libraries in the studio all week.
Actually, the Korg M1 piano ended up on the music of those that had Fairlights and the kitchen sink, simply because it was "the" house/dance music piano of the time. So did many other M1 sounds, actually.
Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...