The TTSDS benchmark evaluates synthetic speech quality by considering prosody, speaker identity, and intelligibility, comparing these factors with real speech and noise datasets. - View it on GitHub