12

I recently bought Garritan Personal Orchestra and find it a great tool to get professional-sounding orchestral renders of my scores. However, I feel a certain lack of "human" playing in it, i.e., chord changes sound a bit "robotic" at times and longer notes are perfectly sustained for long periods of time (they do not decay), which makes the performance sound a bit too computer-generated.

I understand professional musicians and film score makers use such software on a regular basis, yet they obtain much more human sounding results. Is there a trick to making the score sound less "computer-generated"? What can I do to make the results appear closer to human performance?

Improve this question
  • 4
    I'm not sure I can get a full answer out of this comment, but... don't play 'chords' with an orchestra. Record each line separately. It's not as human as 6 humans playing a single line, but it's certainly not as 'blocky' as a keyboard player trying to do the entire string section in one go. – Tetsujin Feb 6 '18 at 19:03
  • @Tetsujin I agree completely - I always separate my short score into individual instruments before rendering. – Klangen Feb 7 '18 at 11:21
  • 1
    I literally play one instrument at a time; I don't work from a 'midi score' of any sort, I play 'freehand', complete with individual controllers per instrument. Each on a separate channel/track so even if I have 3 violin parts, they're all handled separately by the sample-player. I'll also often double a solo violin over the 1st violins' part & drop it back in the mix a bit, to give some separation between the otherwise very similar-sounding samples. EQing each differently can also add to the distinction. – Tetsujin Feb 7 '18 at 11:38
15

I just watched a series of videos of film composer Hans Zimmer who mocks up all his soundtracks using software and virtual instruments so the directors can hear what it's going to sound like before they record it with a real orchestra or other musicians.

He has a control surface with several faders on it, and each fader is mapped to a different MIDI control message that changes a performance parameter on the virtual instrument. So he can call up french horns, hit a chord with his right hand, and then use his left to grab a fader and perform a crescendo on the horns. The other faders change all kinds of things like articulation mode (e.g., change to tremolo and change the speed of the tremolo) and reverb levels, etc.

That requires a few ingredients:

  • Virtual instruments that support MIDI CC messages to make performance changes. Garritan probably supports some of this.
  • Hardware CC controllers to map to the controls in the virtual instrument.
  • The time to map them up in a way that makes sense to you.
  • More time to practice using them to create performances that sound how you want.

A lot of keyboard controllers include at least two built-in performance controls: the mod wheel and aftertouch, so if you have a MIDI keyboard you might not need to spend a lot on external faders or anything like that.

After recording a MIDI performance, or if you've sequenced it from scratch, there are plenty of ways to add randomness and humanization.

Some DAWs (I know Pro Tools does this) allow you to randomize the parameters of a group of selected MIDI events all at once. This is especially effective for velocity.

Beyond that, a very effective but also very time consuming technique is manual editing, informed by an understanding of how the real instruments work and how they are played. Two examples are drums and violins (and other members of the strings section).

When humans play drums, they use both hands, and virtual no drummer has identical hands. Also, when playing a repeated pattern, it's very difficult for a human to not "swing" it slightly. By that I mean, the two hits are not spaced perfectly evenly, usually every other hit is slightly early, and the hit that's on time is usually slightly louder. So for drums, taking every odd hit (the first, third, fifth, etc.) and making it slightly louder, and then taking every even hit (second, fourth, sixth, etc.) and moving it slightly forward in time can create a more realistic sound.

A similar thing applies to bowed string instruments. The upbow sounds different from the downbow. Also it's almost impossible to perform perfect tremolo - the timing will be just ever so slightly skewed. Often the solution there is to use a separate tremolo sample, so changing samples in a performance might be how you can create the articulations you want. The same applies to drums with rolls.

There are plenty of other little details to think about. Enough legato notes played on a clarinet will seem odd because there's no time for the player to breath. Little tiny details that every notices without knowing that they noticed.

Improve this answer
  • 1
    Great answer! When I write music on the computer I always go through by hand and alter all my automation channels like this. For an individual note I even alter the ADSR, reverb, and amplitude in a slightly random fashion to mimic random variations that are always introduced by a human playing in an environment that interacts with sound in a complex way. It's a lot of work but with some pretty good samples you can go pretty far with it. – syntonicC Feb 6 '18 at 17:48
  • @syntonicC Sounds like you need a random LFO that you can use as a midi controller to randomize things automatically. There are DAWs (Pro Tools for sure) that include randomizers for all kinds of MIDI data. In Pro Tools you can select any number of MIDI events, go to operations and randomize any parameters for all of the events at once, with control over how far the parameter can randomly deviate. Maybe I should add that as a humanization technique to the answer. – Todd Wilcox Feb 6 '18 at 17:53
  • 1
    "Enough legato notes played on a clarinet will seem odd because there's no time for the player to breath." Circular breathing :) – user1803551 Feb 6 '18 at 20:45
  • 1
    @Steve I’m writing a major project in musescore right now. But im not expecting musescore to really render a performance from the score. I really just listen to it to see how the melodies and harmonies play. Sometimes I spend extra time on articulation if the exact feel is important. The problem is you can’t really humanize musescore because it doesn’t easily lend itself to making micro level changes. It’s just meant to reflect the macro. The blueprint. You have to build some kind of real or fake performance and musescore is not that tool. – Todd Wilcox Feb 7 '18 at 2:30
  • 1
    To add support on "Some DAWs allow you to randomize the parameters of a group of selected MIDI events all at once.", Mixcraft has a literally "Humanize Notes" function for MIDI notes (disclaimer: not affiliated, just a happy user of it) – Andrew T. Feb 7 '18 at 2:50
2

Is there a trick to making the score sound less "computer-generated"?

Certainly. As you may know, most of the music produced today, including pop songs and movie soundtracks, are computer generated. Only the most famous singers and feature films can afford to hire musicians to record their music, because it is so expensive. The general audience will not realize, although subconsciously they may find the recorded music "better".

You are looking into a profession on its own: music production.

(...) yet they obtain much more human sounding results

The music production process takes years if not decades to master. A 4-minute soundtrack may take a team of 3 engineers a month to produce, using hardware and software that easily add up to hundreds of thousands of dollars. If you are just starting out, do not be disappointed if your soundtrack is not nearly as good as those you would like.

The first step is to get a Digital Audio Workstation software, or DAW for short. It is unclear from your question whether you already have one, or you are just rendering sounds from the score software directly.

The second step is to get a good virtual instrument. A high quality virtual instrument library is very expensive. Garritan Personal Orchestra is good for beginners for its low price and simplicity, however you will not have detailed control over note articulation, breathing, vibrato etc.

No matter what library you use, you need to experiment its full potential. Open up the DAW, fire up the instrument, fiddle the settings then play it on a keyboard. Observe how well you can emulate a human player and produce notes with a variety of tonal characteristics.

(...) longer notes are perfectly sustained for long periods of time (they do not decay)

The third step is to play every instrument, line by line, into the DAW and record the data. This is the core reason why your sound is "robotic" - a real human player will interpret the melodic line and play them musically. If you simply dump the MIDI data from the score software into the DAW, it will always sound robotic. This process will take time, but it will greatly improve your soundtrack quality even if you are using a very cheap virtual instrument. There are "humanization" features in some software, which basically randomly offsets some values, but my opinion is that they don't quite achieve their objective.

For this you would need a MIDI keyboard, a pedal and an ASIO sound card. For entry level options these hardware should be affordable if you can afford GPO.

The next step is to fine tune the performance data recorded. You will need a few hardware fader controls, which is included on most MIDI keyboards. You will at least want to fine tune the "velocity" data of the MIDI, which controls loudness. If your virtual instrument can control vibrato, great. If it can control legato vs staccato, even better. These parameters can either be recorded together with the performance or recorded on a separate run.

This step is optional: If your orchestral score calls for 2 instrument, record them using different virtual instruments. In the real world, it is very unlikely that both flutists are using the exact same flute model. If you "render" the sound directly from the score, the software will just double the volume of the same instrument, which is not realistic. The same principle applies to the entire woodwind and brass section ... I hope you see now why music production is time consuming and expensive.

Then you need to "mix" all your instruments as audio data. In a real orchestra, some players will sit closer to the audience / microphone and some further away. They will have different reverb and left/right balance in the 3D space. When you render the score directly, every instrument is played as if it is right next to the microphone, which is why it sounds unreal.

I hope this answer gives you an idea of how to make your soundtrack less "computer generated" and more lively, and some general directions if you wish to explore further.

Improve this answer
  • It may be possible to get two different flute sounds (for example) by scoring one on pitch and the other a half-step sharp, then de-tuning the sharp flute 100 cents flat. Assuming the instrument is sampled at each pitch, the de-tuning back to the intended pitch will produce sounds that differ from the on-pitch version both by being played from different samples and from passing through the software's pitch-shift algorithm. – Steve Feb 7 '18 at 2:21
  • 1
    To the main point here, I note that playing each line manually definitely humanizes the performance. But it also limits the virtuosity of the performance to that of the person playing each line -- which in my case would sound a lot worse than sounding robotic. It would also give every line the rhythmic feel of the MIDI instrument used to record it, which may be very different from the real instrument; for example, a MIDI keyboard performance doesn't have to take breaths but a wind instrument does. – Steve Feb 7 '18 at 2:27
  • @Steve the pitch-shift method will produce awful sound because they are from the same sound sample. You need a different sample (with a different set of harmonics). give every line the rhythmic feel of the MIDI instrument - you'd use the keyboard to emulate that particular instrument you are playing. If a real player takes a breath and a slight decrescendo at an interval, you do the same. – kevin Feb 7 '18 at 9:58
  • 1
    Thank you for the detailed answer, it really helps understand the whole process better. By the way, I thought GPO was already quite pricey, I was not aware that this is considered "cheap" in the business... :) – Klangen Feb 7 '18 at 11:52
  • 1
    There are many effective driver frameworks for audio interfaces besides ASIO. In fact I don't think you can use ASIO on Macs. Either way, it can just be any interface, it doesn't have to be an "ASIO" interface. @Steve After you've composed and you have a sense of the sound and how it will work, most of the time you'll want a human performance for the final product, if you can get it. Maybe that means hiring a pianist/keyboard player to perform the orchestral parts on a keyboard controller when you can't afford to hire out an orchestra. – Todd Wilcox Feb 7 '18 at 16:41
1

I would recommend 3 steps to humanize the performance.

  1. Use the Instrument Control panels to adjust the settings of each instrument so the timbre is appealing. Some DAW's will also allow you to automate these controls so they can change throughout the track. Check your DAW's manual for more information about automating controls inside a plugin. For example, in Reaper, you can run GPO as a VST plugin, and then draw automation lines to changes the instrument control parameters during the song.

  2. Familiarize yourself with the MIDI controls, especially note velocity. This will prevent the notes from all sounding the same, and allow you to create more human-like dynamics.

https://usermanuals.garritan.com/GPO5/Content/controls.htm#Basic_controls

  1. Manually automate the volume sliders for each channel. This is time consuming, but is a common tool in professional production to create dynamics and balance each section's instrumentation.
Improve this answer

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.