related to the installations themselves through people knocking on the doors as they recorded or scraping the stainless-steel speaker grilles. In this way the work chronicled both its social and physical surrounds. Individuals made creative use of the work adding many nonverbal sounds such as mobile phone tones and the sound of portable CD players.

Technical Notes:
Summoned Voices consists of multiple installations each with a computer and sound system within a chamber at the rear of the door. All computers are connected on a LAN. When the intercom button is pressed the computer records and analyses the sound, saving the audio data as a file and the analysis data to tables in a database on a server computer. The analysis involves segmen- tation of the audio into bursts at various resolutions - corresponding roughly to sentences, phrases and words or parts of words. These segments are then ana- lysed for a variety of characteristics such as the

duration, median pitch, the pitch envelope and spectral charac- teristics. The computer also determines higher-level characteristics such as whether the recording represents: singing, slow speech, fast speech, whistling and so on. Once this data is saved on the server, the local computer then constructs a search strategy. This is performed on the basis of these high-level findings to find matches for each voice segment within the analysis database of previous recordings. It then loads the appropriate portions of audio files from the server into memory and plays the audio through the sound system using a variety of synthesis algorithms. Sum- moned Voices was programmed in C on Linux and used the sound language Pd and the database PostgreSQL along with a variety of scripts and utilities."

(Iain Mott, Marc Raszewski)