README.md

---
title: UTMOS-v2 Estonian MOS
emoji: "🎤"
colorFrom: "purple"
colorTo: "indigo"
sdk: gradio          # <- oluline!
app_file: app.py     # <- oluline!
pinned: false
---



# UTMOS-v2 Objective MOS

This Space computes objective MOS scores (UTMOS-v2) for batches of utterances.

1. **Prepare a CSV**  
   Columns:  
   * `audio` – *relative* path of each wav inside the ZIP you will upload  
   * `method` – (optional) synthesis system name  
   * `MOS`    – (optional) ground-truth subjective score

2. **Zip your WAVs** so that the internal folder hierarchy matches the `audio` paths.

3. **Upload both files**, click **Submit**, wait.  
   Results appear in the table and are downloadable as `predictions.csv`.

> The model here is fine-tuned on Estonian speech data (Utterance level).