Reproduce “Pronouncing Hard Words: Phoneme Adaptation of a Pretrained Conversational Text to Speech Model” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yashish M. Siriwardena, Dinesh K. Gnanasekaran, Tanmay Laud, Shinji Watanabe, Subhabrata Mukherjee, Shanil Puri
To help a voice agent pronounce drug names, the authors adapt pretrained FireRedTTS-2 with ARPAbet phoneme spellings and an optional human pronunciation in an earlier turn. On 7,500 held-out drug-name clips, PHON 60 reaches 84.2% strict correctness without that reference and 90.7% with it. The latter uses curated recordings; this does not establish clinical safety or transfer beyond an English/ARPAbet backbone.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.