1

I know that in modern sound-production software it is possible to reproduce all sorts of music instruments. But I don't remember that I've seen a software reproduction of a classical guitar.

Is it possible to create a guitar sound in software, which would sound as a real guitar, like in this track, for example?
It is electronic music, but I don't believe that the guitar sound there was created with software.

  • That depends entirely on what you want the simulated "guitar" to sound like... – LSM07 Mar 21 '18 at 13:52
  • 2
    Strange but true anecdote. One friend had used a sample guitar to play a part he later wished to replace with the real thing. Other friend comes over, we mic up, he plays the part. Utter confusion... do we have the mutes wrong, why aren't we hearing the new part only the guide sample?... took a while to figure out... guitarist & equipment were the same guy/gear as made the sample set. Grins all round :) – Tetsujin Mar 21 '18 at 14:00
  • 1
    Are you talking about samples or synthesis? Most of the "reproduced" instruments you are hearing are just multi-sampled and well-articulated. – Todd Wilcox Mar 21 '18 at 14:52
  • 3
    As a software engineer, my answer is that you cannot reproduce an analog sound with a digital substitute perfectly. However, you can reproduce it well enough that no human ear is able to distinguish between the two. – Don Branson Mar 22 '18 at 1:55
  • 1
    @DonBranson nailed it. – Carl Witthoft Mar 22 '18 at 11:51
6

(I think the sound in the song you linked is a steel-string guitar, if it's the sound my ear picked out, rather than a classical guitar.)

The most common way to reproduce a guitar sound with other technology is to use sample-based synthesis to replicate the guitar sound. Whether that counts as a "software reproduction" of a guitar by your criteria, you'll have to tell me :). It's very easy to get good representations of some instruments with this technique, but guitars are a little more challenging than you might think, due to a number of subtleties in, and combinations of, ways you can fret the strings and articulate notes. Modern sample libraries and sample-based instruments often deal with this simply by sampling many different instances of differently-struck notes, including samples that work well as part of a strum, and also fret noises and other playing sounds that make the part sound more authentic.

Here's a sample-based classical guitar:

and a steel string:

In both of those 'performances'', you can probably hear aspects that are convincing, and aspects that are less so. That isn't to say that it wouldn't be possible to do an even better job, but it comes down to the amount of effort that you would need to put into each track to fine-tune it to be really convincing - at some point, if realism is your aim, it's a lot easier just to find a guitarist who can play the piece for you.

It would also be possible to produce a realistic job using physical modelling or additive synthesis - this might more fully fit the brief of being a "software reproduction" of a guitar. However, I don't know of commercial products using these synthesis methods that go as far as commercial sample-based products in handling the many special cases of guitar playing techniques - you'd probably have even more work to do to get to a certain level of realism.

|improve this answer|||||
  • It's all the "golden ear" controversy :-) – Carl Witthoft Mar 22 '18 at 11:51
  • 1
    @CarlWitthoft I wasn't aware there was a controversy about golden ears. Only someone who can't believe that another person might have better hearing than they could disbelieve in golden ears. I know I hear better than many and worse than some, so I'm sure many hear things that I do not, just as many do not hear what I hear. I've sat in a room with engineers and musicians and said, "can we please stop that high pitched feedback?" To which everyone else responded that they did not know what I was talking about. Totally off topic, of course. – Todd Wilcox Mar 24 '18 at 5:59
2

"true"? No. Truth does not depend on quality.

But listen to about 35:15 in this part of a multitrack demo session hacked together within a lecture and you'll find that the guitar track, a pure Midi rendition by Fluidsynth from the output of the music typesetter LilyPond (so with a very very rigid timing) is the least of the problems here.

Now Fluidsynth uses freely available audio samples. Commercial Midi expanders tend to be noticeably better in quality. At home, I'd rather use the output of a Solton MS40 and that thing is from the 90s.

If you are playing tricky solos, Midi expanders will have a bit of a problem delivering them believably. But much much more so with continuous-tone instruments under constant pliable player control like a solo violin (violin sections actually work better). Wind instruments like saxophones are already less conspicuous even though they are ostensibly continuous-control instruments as well. Basically percussive instruments like piano and guitar can be pretty well represented by a limited number of samples and some envelope trickery.

You never can reach "true" by definition, but "good enough" is easily doable for a lot of purposes even with old technology, and certainly more so with newer devices.

|improve this answer|||||
0

I will assume you're asking about synthesis/simulation, and not just making a digital recording and playing that back. Most people can distinguish between recorded sound and being in the same space as a real live instrument, the sound is being created by different mechanisms, sort of like comparing 2D with 3D.

For synthesis, if somebody did enough research, analysis and mathematics, there's no reason why they couldn't create a dedicated simulation of any instrument that didn't use real audio samples.

Reproduction of the sound is not a challenge for the computer, or the software, the challenge is determining the mathematics required to generate the sounds.

Some instruments will need a handful of parameters to generate a convincing tone, other instruments may need significantly more parameters.

Some of those parameters will be fixed because they are connected to the style or state of instrument being simulated, other parameters will be variable because they are determined by the way the instrument is being played.

So I think yes, software can produce a convincing digital reproduction of the sound of any instrument, indistinguishable from a digital capture, if only we solved the problem of how to actually do that.

None of the images in this page (https://www.flickr.com/photos/bbb3viz/) are photographs, they're all generated by a 3D rendering engine. If software can take a 3D model of a landscape and produce photo-realistic images. Then yes, software can absolutely take a model of a musical instrument, and produce the equivalent audio-realism.

We just need to learn how to model them. Something I wouldn't mind experimenting with myself, if only I had the time. :(

As for actually performing that instrument in a way a human would, well that's a whole new level. One step at a time, eh?

|improve this answer|||||
  • Although, the modellers making these models do use real artefacts to generate textures. This could be compared to using samples of real instruments, but I'm sure that's just a shortcut to eliminate the difficulty/effort/time involved in simulating textures. – Lee Kowalkowski Mar 22 '18 at 16:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.