Invitation to try RVCBench: voice-cloning utility, robustness and automated evaluation

[Nanboy-Ronan]

Hi F5-TTS team,

Thank you for developing F5-TTS and sharing your models and code!

I'm Ruinan, an author of RVCBench (accepted at NeurIPS 2026). We'd love to invite you to try our dataset with F5-TTS:

• Conventional utility: speaker similarity, content accuracy, speech quality, intelligibility and emotion consistency.
• Additional robustness dimensions: noisy/overlapping references, compression, anti-cloning protection, long-form speech and unusual text, with speaker-demographic and multilingual/cross-lingual coverage.
• Automated evaluation pip package: pip install "rvcbench[eval]" — seven automatic metrics, prompt export and reports; also usable on your own audio.

For F5-TTS users, prompt exports include a Seed-TTS-style batch list to help connect existing inference workflows.

Hugging Face dataset · Quick start · PyPI

A 52-utterance onboarding suite is available (packaged suites are currently previews). We'd be happy to help with setup or a relevant subset, and would appreciate your feedback.

Thank you for considering it!
Ruinan, on behalf of the RVCBench team