← All benchmarks

Seed-TTS-eval English (UTMOSv2)

Multimodal

Text-to-speech naturalness on the English (Common Voice) test set of ByteDance's Seed-TTS-eval, scored as the mean UTMOSv2-predicted MOS of the synthesized utterances on a 1-5 scale. UTMOSv2 is an automatic MOS predictor, not human rating, and the voice-prompting setup (zero-shot cloning or a preset voice) is a run condition recorded on each score.

Top models (higher is better)