1 paper
Ryota Komatsu, Kota Kawakita, Takuma Okamoto +1
Unsupervised syllabic tokenization aims to learn discrete syllabic tokens that capture latent linguistic content-related structure from raw speech. Recent syllabic tokenization met…