---
title: Italian Speech To Text
emoji: 🏃
colorFrom: yellow
colorTo: pink
sdk: gradio
sdk_version: 6.12.0
app_file: app.py
pinned: false
short_description: Italian ASR + English translation (Whisper + Marian). Optional local fine-tuned model in models/whisper_finetuned_it/.
---

# Italian → Italian + English

Speak **Italian**; get **Italian transcription** and **English** translation.

- Default: `openai/whisper-small` + `Helsinki-NLP/opus-mt-it-en`
- Optional: copy your fine-tuned Whisper into `models/whisper_finetuned_it/` and set `ASR_REALTIME_MODE=finetuned` in Space variables.

## Add your model

1. Copy training outputs into `models/whisper_finetuned_it/` (`config.json`, tokenizer files, weights).
2. Use **Git LFS** for large weight files when pushing this Space.
3. In Space **Settings → Repository variables**: `ASR_REALTIME_MODE` = `finetuned`.

## Optional environment variables

| Variable | Default | Meaning |
|----------|---------|---------|
| `ASR_WHISPER_MODEL` | `openai/whisper-small` | HF Whisper id if not using local finetuned |
| `ASR_REALTIME_MODE` | `quality` | `quality` = hub Whisper; `finetuned` = load `models/whisper_finetuned_it` |
| `ASR_WHISPER_FINETUNED_DIR` | (see above) | Override path to finetuned folder |
| `ASR_TRANSLATE` | `1` | Set `0` to disable English translation |
| `ASR_MIN_RMS` | `0.005` | Silence gate |

## Local test

```bash
cd Italian-Speech-to-Text
python app.py
```

Configuration reference: https://huggingface.co/docs/hub/spaces-config-reference