Prosody

Prompting the Muse: Generating Prosodically-Correct Latin Speech with Large Language Models
Prompting the Muse: Generating Prosodically-Correct Latin Speech with Large Language Models

I will present a talk exploring how historical linguistics can partner with AI, presenting a case study on Latin prosody and the use of large language models to generate metrically correct speech. This presentation is part of the /"AI and Digital Humanities/" conference at the Università di Siena.

Nov 5, 2025

Veras Audire et Reddere Voces — Prosodically‑Accurate Latin Text-to-Speech Corpus and Methodology
Veras Audire et Reddere Voces — Prosodically‑Accurate Latin Text-to-Speech Corpus and Methodology

An openly licensed, expertly validated corpus of Latin poetic audio that makes classical metre audible with modern large‑language‑model TTS. Drawing on Pedecerto’s machine‑readable scansions, we steered a general‑purpose model to recite 216 lines—Vergil Aeneid 1.1–100 (hexameter) and Ovid Heroides 1.1–116 (elegiac couplets)—with metrically faithful stress, yielding nearly 24 minutes of line‑aligned, classroom‑ready audio. This project was presented at the ACL 2025 Student Research Workshop (Vienna) and at CLiC‑it 2025 (Cagliari).

Jul 26, 2025