oliveros bot
Arts
Electronic Music Minor
Pauline Oliveros was a queer experimental musician whose work focused on breathing and listening. Her work includes a large number of text scores which rarely include traditional western staff notation. They are instead a set of instructions, often poetic, intended to create a specific experience through the artist’s interpretation. This seemed perfect to use to train a text generator.
I used the GPT-2 language model, trained on Oliveros’ Anthology of Text Scores (2013) to generate text. I trained the model a number of times, saved the output, and then retrained it with slightly different parameters and/or source. I did this 12 or so times, saving the output for each. From this selection I chose pieces I could perform relatively easily solo, and a few interesting poetic ones. These are the basis for the performance video. The video of the performances is narrated by Real Time Voice Cloning: a text-to-speech synthesizer that can be trained on a voice source, using SV2TTS. The source is a video titled “Deep Listening Certificate session with Pauline Oliveros” from 2010.
The video has seven pieces, which are clearly described. Four of them are performed. The order and some of the editing was chosen for emotional impact, but the intent of the video is to be a series of experiments.