Wavenet-for-Poetry-Generation [REHEARSAL]
An improvised reading,
of generated poetry made as it is born,
at 201701110-1347.
Practice runs
[TEXT-to-Human-VOICE & VOICE-to-TEXT EXPERIMENTS]
Preparing for an installation exhibition-performance where viewers will be allowed to speak generative texts flowing over walls. All speech will be converted into text.
1. Training the voice reflexes to hurtle over the avalanche of words.
2. Using Google docs in fullscreen with text-to-speech. Severe latency. And confusion as the language models on Google’s servers try to construct something normal from the erratic non-linearity of poetic speech. Txt-to-speech is overfit for normal thought.
3. Using Siri in Sublime text. Again: freezing, miscomprehension, latency, confusion. Another solution must be found.
4. Made in cold kitchen under big heater. Too fast.
O how broken and full it all is.
