1 paper
Mizbaul Haque Maruf, Muhammad Nur Yanhaona
Text-to-speech (TTS) corpora are costly to record, yet many utterances add little new phonetic information. Core-set selection reduces this cost by choosing a small training subset…