Max was used both for the preliminary sound design, as well as for the installation.
A first max patch was created to design each imaginary call. Each call is made with a combination of breathing and vocal tract sounds. The breathing sounds were made out of air pumps recordings played back through granular synthesis, with some controlled randomness in the playback for a more lively rendering.
The synthesis for the voiced part of the sound uses a port of a vocal tract model by Andy Farnell, originally coded in Pure Data. A significant effort consisted in creating control curves for the model, allowing to interpolate smoothly between vocal tract presets. The core of the curves model is done with the "prob" object combined with Jasch's "bspline" object. This helped us derive hundreds of controlled variations from each seed sound.
All of the sound variations were encoded and stored in "coll" object, and are resynthesized by the installation patch, which plays them with a random process for a cycle length that can can be set to last between 2 mins up to 2 hours, depending on the installation's constraints.
The installation patch has been running with Max-6 runtime every day for the past five months without a complain, only plenty of groans.
Once the exhibition is over, we are thinking about releasing the sound-design patch as an instrument for electronic music producers and see what kind of remixes can roar out of this journey.