Poeta Ex Machina! How To Teach Computers To Chant Latin Poetry!
summary
Claims a working Latin text-to-speech that actually chants scanned meter, not just a demo. It uses concatenative synthesis with a roughly 10 MB voice database called umbrella, pulls scansion and forms from Perseus-style HTML/APIs, models pitch from breath contours, and implements glue logic in about awk 300 lines of code with caching that once created millions of tiny files. The live web app at poetaexmachina.net mostly works and plays multiple synthesis backends and audio examples.
AI-written summary · about
chapters (5)
- 0:00Start
- 2:01Speech synthesis: pink trombone!
- 3:56Speech synthesis lots of voice data!
- 5:51Sample input!
- 10:02Latin Grammar!
related talks
11mMaking blackout poetry with computers!
33mFor AI, By AI: Freedom & Evolution of the AlgoPoetic Avant-Garde
42mBeats, Rhymes and Unit Tests
05mInterview
27mBuilding a Low-Latency Voice Assistant leveraging Elixir and Membrane
32mBeats, Rhymes and Unit Tests
48mYour Apps Can Talk! Introducing Cloud Text-to-Speech, Powered by WaveNet Technology
16mServing Voice AI at $1/hr: Open-source, LoRAs, Latency, Load Balancing