---
title: S2ST Subjective Test
emoji: 🎧
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 5.16.0
python_version: '3.10'
app_file: app.py
pinned: false
---

Anonymized listening test for speech-to-speech translation models. Each
sample shows 9 anonymous model outputs (Model A..I, randomly shuffled per
sample per rater). Raters score each on three 1-5 axes: translation quality,
audio naturalness, speaker similarity. Submissions land in `mos_results.csv`
(ephemeral - download regularly from the Admin panel).