I began this project by researching John Chowning's vocal synthesis technique that was used in his 1980 piece Phoné.
I was originally interested in creating a real-time performance device that would create multiple voices based on audio input
(presumably a human singer). Eventually, however, I decided that it would make more sense to create a pre-recorded chorus with
ChucK generated voices in order to have more control over the finished product. I decided to use a choral work from the 15th
century because vocal music of that period tended to focus less on text and more on clarity of voice and harmony. Because the
vocal synthesis program currently cannot enunciate consonants, I decided that I should probably choose a piece that does not
involve a lot of text.
At first I had a lot of problems making the voices sound authentic. Eventually, Chris helped me get the vibrato down to a point
that sounded a lot better than previous versions. I had also planned to have just the synthesized voices singing the piece, but I
decided that adding real voices would help disguise the limitations of the computer voices. I recorded my friends Eric and Will
each singing two parts: Eric on soprano and alto and Will on tenor and bass. It took a lot of editing to get the computerized
voices to align with the real voices. In fact, this project has taught me a lot about tempo and how humans perceive beats versus
a computer. Because the vocal synthesis program relies on small portamenti to avoid clicking and to sound realistic, anything
less than a half note was totally lost, which meant that I had to chop up huge portions of the ChucK voices to get them to match
up. In fact, I decided to eliminate the ChucK voices entirely in faster sections, especially the Christe section.
After finally getting the voices to align, I got a MATLAB script from Jonathan Abel that generates impulse responses. Using his
script, I could customize the intensity and decay time of each critical band. I put each voice through this script, I spent a
long time manipulating the critical bands to try to create an effect of spatialization, which was a pretty inexact task. While
the resulting mix sounded pretty good on headphones, it did not take advantage of the 8-channel ring that was set up in the back.
The day of the concert, Chris and I decided to re-do the mix on Ardour so that we could use the 8-channel ring. We added Freeverb
to each track, slightly varying the width of the rooms for each voice. We ended up turning the ChucK tracks down pretty far, so
that they became textural additions. I was very happy with how it all turned out and thought that it was especially effective in
the outdoor setting.
About Missa Prolationum
The Missa Prolationum is a work from the 15th century that is made up of overlapping mensuration canons (a canon in which the voices
sing not only the same melody, but at different speeds). The Kyrie section features the four voices in pairs of mensuration canons on
the unison. Throughout the Mass, the canons becaome intervallically further apart. Each voice sings in a different time signature,
the superius in common time, the alto in circle time, the tenor in imperfect cut time, and the bass in imperfect circle.
I spent the first two weeks of the quarter learning a bit about John Chowning's vocal synthesizer as first utilized in his
piece "Phoné" and Chris' own edits to the code that he has made in recent months. It's been very interesting exploring
the ways in which a computer can and cannot simulate the various sounds produced by the human voice. I hope to improve some
of the vowel formants (particularly the open o sound). To that end, I've been consulting my voice teacher and have been looking
for newer, more accurate formant charts.
Eventually, I would like to create a real-time performance tool for my final project -- a sort of "make your own choir"
deal, but first I will focus on improving the code that currently exists.
Chris' FMV executable is located here on the CCRMA network, in case anyone wants to play with it: /user/c/cc/Desktop/dsp/ccrmaFaust/fmv
Also, special shout out to Chris for his awesome code and to Spencer for making sure it works on my computer.
I've primarily been focusing on reading papers on different methods of vocal synthesis, as well as papers on the acoustics of
the voice. I've been especially interested in Xavier Rodet's CHANT project, which has some very interesting observations
on how to accurately synthesize vibrato (i.e. jitter in the form of unescapable pitch fluctation of the fundamental). I have
also been trying to understand more of how John Chowning's vocal synthesis works -- I'd very much like to get a hold of his
paper on the subject, although I keep forgetting to ask Chris for a copy. I've also been playing around with the code that
Chris sent me; currently I have it singing row row row your boat (albeit poorly), which is pretty entertaining. I was thinking that
for the final project, I would like to focus on getting these vowels to sound as accurate as possible and to get this code as
functional as possible; instead of a live performance, I was thinking that I might try to write a code for these Chowning voices
to sing a choir piece together. I'm thinking something by Josquin or Ockeghem because their music is methodical and also
there is more focus on the vowels instead of the consonants (just in case I don't get the consonants up and running).
here is my ChucK
code currently, although it will not run on the version of ChucK that is currently public.