Spaces:

---

Navid-AI

/

Arabic-TTS-Arena

Like 67

Running

Hello Arabic TTS Arena maintainers,

We are validating an audio-language model as a judge of same-text TTS pairs. Do you provide a downloadable, anonymized subset of the Arena's Modern Standard Arabic votes with the exact audio clips that listeners rated?

We need at least 50 distinct MSA pairs, original text, system identifiers, audio IDs or hashes, preference/tie labels, and anonymous rater/session identifiers where available. If several people rated the exact same pair, their individual votes would help us select clearer preferences. We would keep dialects separate and exclude natural-vs-synthetic comparisons.

The public leaderboard alone is not sufficient for this purpose, and re-synthesizing text would not preserve the identity of the evaluated audio. Could you advise on access and the permission terms for local internal evaluation and any derived manifest? We do not require personal information about voters.

Thank you.