I was a user of the NED Synclavier II in the late 1980s. My dissertation piece, Cenotaph, was
created using the resynthesis capabilities of the system.
After graduation, I lost access to the Synclavier and pursued music by other means.
Now, 30 years later, enter the Synclavier V. I feel reborn. Thank you, Cameron Jones and Arturia!
In the old days, resynthesis was a somewhat laborious process. The user had to input the starting point
and ending point for each timbre frame. It was advisable to place the markers at zero crossings. It was
tedious and very much constituted an art in itself, which required a lot of practice and trial and error to get
right.
Resynthesis in the Synclavier V has been streamlined considerably. Placement of the frame markers manually
is vastly easier than in the old Synclavier II. (Don't rely on the automatic frame placement for anything with
any complexity.Manual placement is best.)
I'd like to discuss some tips for achieving a good resynthesis. Since my primary interest has been the
resynthesis of the spoken word, the tips will be primarily focused on that, though I don't know why they
could not be more generally applicable.
But first, an example:
Here is an mp3 of an original wavefile in which I am speaking the name of the Aztec god, Huitzilopochtli.
(I like the Aztec god names because they are rich in many different phonemes):
see attachment 1 (Huizilopochtli.mp3)
Here is a recording of the resynthesized result:
see attachment 2 (HuitzilResynth.mp3)
To my ears, the resynthesis is absolutely remarkable in its fidelity to the original wave recording.
I can actually recognize my own voice in the resynthesis. Notice that the consonants come through clearly.
The monotone of the resynthesis is presumably a byproduct of the new resynthesis engine. In the resynthesis
engine of the Synclavier II, the pitch content of the frames was preserved because the user input both the starting
point and ending point for each frame, thus defining the duration in samples of the frame. That essentially defined
the pitch content of each frame. Each frame would then have a "pitch offset" associated with it.
The newer engine determines those starting and ending points on its own, so it calculates the waveform based on the autotune
position of the wave file that is being resynthesized.
Before loading a sample for resynthesis, I load the "Template: Default" timbre. This will be the baseline for a clean
resynthesis. You can also choose complex timbres as the baseline for different effects. That's one of the beauties of the
Synclavier V: it is enormously flexible.
Tip One: It is important to choose a good autotune analysis position.
I find that when loading a sample for the first time, the autotune position is always 0.3, and the default
key setting is C3. You can move the autotune position easily. I find that I get the best results in resynthesis of my
own voice when the resulting key after autotune analysis is applied is somewhere around F#1. You will need to experiment
with the position based upon the wave sample you are resynthesizing. Autotune is discussed in the manual, and it is stated
that it can make "all the difference" in the result. I find this to be very true.
Now, how does one choose where to place the frame markers?
I have developed an approach which seems to work quite well on the voice samples I use. Basically, it boils down to
placing a marker at every point where there is a change in the waveform.
For vowels, this amounts to placing a marker at the onset of the vowel, the rise of the vowel, and then enough markers
to capture variations along the way.
Consonants have a much larger noise component, so care should taken to capture every transient change within reason.
DO NOT BE AFRAID TO ADD FRAMES. More is not always better, but very often it is necessary for a truer resynthesis of the
consonant sound.
For example, I used 51 frames for "Huitzilopochtli." In fact, that is at the lower end of number of frames that I have used for
similar words.
Here is Mictlancihuatl:
wave recording (mp3); see attachment 3 (Mictlancihuatl.mp3)
resynthesis: see attachment 4 (MictlanResynth.mp3)
I used 73 frames for that one. I essentially nailed it on the first try by following the simple guidelines outlined above.
(I've also had a lot of practice while developing the procedure.)
I have found the best results in placing the marker as near as possible to the peak of a waveform,
rather than near a zero point. Expand the scale of the timeline so that you can see the waveform in order to
make the placement more precise, though the engine is actually rather forgiving in that respect, I find.
After the frames are placed, "Apply" the resynthesis. Choose "Carrier" or "Both", depending on what you intend to to do with it.
Test the result. If you don't like it, it's extremely easy to reload and try again.
To sum up;
1. Choose the autotune position carefully and experiment with different placements.
2. Don't be afraid to use as many frames as necessary to capture the important transient points of the sound. You can always
delete frames at will.
I take the resynthesized items described above and greatly extend the frame splice times so that the original name is unrecognizable and becomes a long duration sound with an evolving timbre. I then stack and manipulate these in various ways to produce what I think are interesting results.
I have recently uploaded an album of new pieces using the resynthesis of Aztec gods' names as outlined above for the entire sound content, excapt for the last track. Included in the album as the last track is my piece, Cenotaph, from 1989 using the old original NED Synclavier II system and its resynthesis capabilities. It's sound sources are the resynthesized names of the astronauts killed in the
Challenger space shuttle disaster.
If you are interested, here is the link to my album on Bandcamp:
https://rownan.bandcamp.com/releasesHave fun and Happy Resynthesizing!
(I have no idea how the attached examples will show up. I hope they are here somewhere.)
