posted 23 June 2003 00:29            

He is my wife's uncle, and so I have special access to him. He sent me his score for the 5 part trio, and I synthesized and rendered a recording of the second movement this past week.

Along the way I have a number of obstacles to overcome. I was introduced to the subtler details of sampling synthesis and find that even the very best sample libraries lack solo Violins and Cellos. [Although this is promised in the Pro edition of the Vienna Orchestral Cube -- for a mere $5,500 USD!]

So I had to make use with whatever sound scraps I had on hand. The Cello was especially challenging because this movement depends crucially on octave bowing transitions from a bass note with the string held against the string board, and the upper tone bowed as a harmonic -- i.e., lightly touched string.

The violin was especially maddening because, despite my samples sounding excellent in isolation, when mixed with the other sounds it always sounded more like a harmonica to me. I did the usual EQ tricks of scooping out space for the instruments in their special regions among the other instruments. That helped to emphasize the different instruments, but the violin still sounded like a harmonica at pianisimo dynamic levels.

So.... I broke out my copy of Arthur Benade's "Fundamentals of Musical Acoustics" and in chapter 24 he describes the physics of instruments in the violin family. After reading carefully about his measurements, and how sound is produced by these instruments, I began to feel hopeless of ever rendering a reasonable facsimile of a live violin or cello.

But then an inspiration struck me. Visual artists routinely use slight exaggeration of certain colors in the midst of surrounding context colors. For example, a sunset scene will feature traces of flourescent orange in the midst of blues and purples. Used sparingly the effect can be quite striking. Overly used, it begins to resemble a child's painting.

So why not try something like this to bring out the special character of the violin and cello when layered with other instruments. I took Benade's prescriptions for the 1st air and wood resonances, as well as the broad bridge resonance region and applied a touch of these to sounds that should already possess this character. Using very slight EQ levels, I exaggerated the known resonance characteristics of these two instruments. In isolation, it would be too much even at these low levels. But when layered, this was just the magic needed to make them retain their distinctive sound qualities.

Here is the prescription I used:

Violin...
1. parametric peaking filter at 290 Hz, +6 dB, 0.1 octave bandwidth, for the first air resonance
2. parametric peaking filter at 440 Hz, +6.5 dB, 0.1 octave bandwidth, for the first wood resonance
3. parametric peaking filter at 3000 Hz, +3 dB, 1 octave bandwidth, for the bridge resonance
4. highpass filter at 158 Hz, Q = 0.71 (very shallow)
5. lowpass filter at 6337 Hz, Q = 0.71

Cello...
1. parametric peaking filter at 125 Hz, +6 dB, 0.1 octave bandwidth, for the first air resonance
2. parametric peaking filter at 175 Hz, +6.5 dB, 0.1 octave bandwidth, for the first wood resonance
3. parametric peaking filter used as a deep notch at 1500 Hz, -6 dB, -0.1 octave bandwidth
4. parametric peaking filter at 2000 Hz, +3 dB, 1 octave bandwidth, for the bridge resonance effects, in combination with that notch at 1500 Hz.
5. High-pass filter at 72 Hz, Q = 0.71
6. Low-pass filter at 6469 Hz, Q = 0.71

All of these filters in combination produce a spectral response with a shape similar to the special "formants" of each instrument, but the amplitudes were intentionally kept relatively small. All I needed was the subtlest exaggeration.

The results sound fabulous. I was finally happy enough with the rendering to cut a CD recording of it, for audition by the composer.

[It strikes me that this little trick might well be already known to the professional audio engineers out there. Since I'm not one, I'll freely admit to this technique. No doubt some expert has this little goodie squirreled away in his little black notebook... eh? Pete? Mathis?]

Cheers,

posted 23 June 2003 07:21            

My little black book has a lot of empty pages in this aria which I am starting to fill up with ideas, mostly inspired by your postings. Samplers are very limited in what they can do but normally the problem is with exposed and solo sounds. With strings I can normally get a reasonable sound if it is emulating a large string section (washed in reverb and EQed to death) and is buried behind some thing else. For solo violins and cellos the only way I've found of getting it to sound reasonable is to limit what I play. i.e. get the sample first and then compose what sounds good with that sample. This is very frustrating and I'm hoping that Physical modeling will provide the answer.

Some time back you posted a cello using Karplus-Strong which had a sound I had never heard before with a sampler or synth, where you could hear the string ring on after the bow had been removed. You could also get the scraping sound as if the bow was being moved too slowly across the string. I could only get the sound correct for one note at a time and had to readjust the sliders every time I played a different note. I hope to be going back to try to improve on this.

You also more recently posted some attempts at Physical modeling . I got some very violiny sounds out of this when I put the feedback up high but it I think it needs a low pass filter in the feedback path to get rid of the artifacts building up at high frequencies. Once again I need to revisit this when I've made a multi resonator (the kyma resonator doesn't have a hot parameter for the brightness control).

Now you've given me another trick which might help out with the ultimate string synth.

One thing that adds to making string samples sound like a harmonica is that the sample is re triggered on every note played. As a harmonica has a separate resonator for every note played its like having 30 separate instruments in one package. Each instrument has to got through it's startup phase before it makes a sound , even if the player is blowing continuously.

So if instead you were to record each violin/cello part separately as a non poly type sampler, you can make the sample player work such that the samples pitch follows the keyboard tonally, but gets retriggered only when all of your fingers have been removed from the keyboard and a new note is then played. This is called autophrase and you very quickly learn a different way of playing the keyboard. Also if the violin sample has vibrato later on in it's cycle, it will still be heard even when fast moving phrases are played. You can also use portamento to smooth the transitions but this has to be set to automatically switch off when the sample is re triggered. Here the note on velocity of all but the first note in the phrase can control the portamento rate. Also note on velocity of the first note in a phrase can control the attack time of an envelope and note off velocity can control the release time. This also becomes quite instinctive with your keyboard playing technique.

Anyway
in answer to your question, is your technique in my little black book , well it is now.

Thanks

posted 23 June 2003 10:53            

Gee, thanks for all that wonderful feedback! I was also thinking of trying this filtering technique to my physical modeling synths. But you right now tickled my imagination into how to construct the autophrasing you mentioned, and controlling envelopes, portamentos, and vibrato with first notes. Surely Kyma can do this sort of thing, but I have to think about how...

Cheers!

- DM

[BTW -- I had to work like the dickens to get a reasonable approximation to the violin. Before I discovered this filtering trick, I found an almost perfect sample on the Proteus 2000, complete with fingered portamento and bowing noise. But alas, a few low-velocity samples above C5 were out of tune and unstable in pitch. It sounded just fine in solo, but when combined with the piano and cello, the tuning was a major killer here. Apart from sounding flat, it also beat horribly with the other tones and produced a highly offensive result in the pianissimo sections.

I ended up backing off and using a string section sound, heavily lowpass filtered above 3 KHz, and then feeding this exaggeration filter. The result is not quite correct, but at least it is more correct than anything else I have here. The problems all show up at the highest pitches above G5, where it does sound a bit like a string section instead of a solo violin. The high partials are very unstable in time. Physical modeling may well be the answer at these high pitches...]

posted 24 June 2003 01:11            

I'll check out Lancaster's article. It sounds pretty interesting. I used to follow him many many moons ago when personal computers only had about 256 bytes of RAM...

Gee John, we were both there and we didn't even realize it at the time! So you heard the Trio that I am rendering. I felt bad because at the time the music was so complex and just zipping past my head, and there was no way I could hold onto and study it in that short time. Having the score now in my hands allows me the opportunity to study and learn from Seymour. I was quite taken by the 2nd movement when the Cellist was performing harmonic octave bowings. I felt like Seymour had written that part for the physicist in me...

I am just now thinking about rendering the music totally synthetically for own amusement. I did it as accurately as possible with realistic instruments for Seymour's sake. He, to this day, still does all his composing at a piano using pen and ink. We are slowly introducing him to Sibelius, but I fear this quality of synthesis will elude him, for lack of his technical understanding of sound and sound production.

As for GA synthesis! Yes, indeed I tried that with some degree of success on the Cello. But there were a couple of samples of low string bowings that had an interesting shift of emphasis on higher partials about halfway through the sample. I could not get the spectral analysis to recognize that transition properly. It kept showing up as a scraping noise in the resynthesis.

I had originally hoped to create a morphing instrument like the sample Trombone in the Kyma Library. Perhaps I need to revisit this particular problem and work a bit harder to get the non-pitched sounds and that transition to appear properly. I still think that GA synthesis on top of a good spectral analysis would lend itself to a more flexible instrument.

Cheers, everybody!

- DM

[I will say this much... I first got introduced to sampling synthesis back in the mid-80's, first on an Emu Emulator-I, and then on my own Mirage, as part of the effort to automate a video mixing studio for documentary films. (a very early effort at what we all take for granted now...)

I gradually developed the notion that it was a waste of time to attempt synthesizing realistic live insttruments. And I too, was a fan of, and read everything I could from Wendy Carlos. So I have had my back turned on this aspect of synthesis for nearly 2 decades now.

However, now that I have a lay person and composer as an audience, I felt it would be better to approach him as realistically as possible. Hence getting back into synthesis of live instruments has been an eye opening education for me these past weeks. Not only is it difficult to do such a thing, it is darn near impossible.

But I am impressed by the demos offered by the Vienna Symphony and their Orchestral Cube. Another impressive collection of samples comes from Sonic Implants, and even the Halion Strings from Steinberg. All of these demonstrate breathtaking illustrations of sampling technology. But what I find interesting are the things they take care NOT to demonstrate.

So I take to heart what Pete was saying above. There are only certain kinds of musical passages that can be rendered with samples, and not others.

Sampling has come a long way in 20 years, but my own preferences for synthesis are still for synthetic sounds that can't be heard any other way. GA synthesis offers a possible middle ground here. ]

posted 24 June 2003 04:37            

On top of the cello and violin sounds I imposed that EQ prescription at the top of this topic, then I scooped out room for each sound in the spectrum using shelving filters only 3 dB deep over their respective principal regions.

The piano and cello go through a pair of tempo sync'd feedback delays. So the sound is a bit jazzier than originally performed.

I hear a bit of Bach style in the Allegro portion in the middle of this piece, but the ending is pure Barab. Sadly, this rendering can't do justice to the poignant beauty of the final 40 measures...

"Switched On Barab" at http://dmcclain1.home.mindspring.com/SynthMinstrelBoy.mp3

Cheers,