speaker-representation

SPARCLE is a speaker-aware grapheme representation model that uses contrastive learning to align grapheme embeddings with acoustic representations, improving text-to-speech quality especially in low-resource settings.