Leveraging Book NLU for Enhanced AI Narration

In recent years, audiobook creation has undergone significant transformation. Advances in text-to-speech technology now enable highly expressive, human-like narration with impressive quality, democratizing the storytelling process. Audiobook production remains a human-in-the-loop process. Could book Natural Language Understanding (NLU) further assist creators in enhancing this workflow?

In this talk, I will explore several research efforts aimed at helping with the audiobook creation through Book NLU. I will cover key areas such as quotation attribution, speaker identification, and character modeling in novels. Additionally, I will highlight the critical role of memorization in large language models (LLMs) and argue that evaluation for these systems should take it into account for more accurate performance assessment.

While these key areas are showcased in the context of enhancing AI narration, they hold broader significance in the field of digital humanities. Techniques such as quotation attribution and character modeling are equally valuable for analyzing large literary corpora in order to get insights into texts at scale.