Speech synthesis and voice cloning
Seed-TTS-eval
Zero-shot TTS/voice conversion in English and Mandarin using reference audio.
About Seed-TTS-eval
What it measures
WER and speaker-similarity cosine score under the official ASR/embedding stack.
Scope & limitations
Full protocol requires compatible reference clips and permitted cloning access. Deepgram/WellSaid stock voices cannot be assigned an equivalent speaker-cloning score; stock-voice WER is a separate adaptation.
Evaluation environment
Reference-conditioned TTS API; generate WAV/audio for the prescribed text.