Datasets:
| language: | |
| - ar | |
| task_categories: | |
| - text-to-speech | |
| pretty_name: Mimir Auto Tokenized SNAC | |
| # Mimir Auto Tokenized SNAC | |
| Tokenized TTS training dataset derived from the paired audio/text dataset. | |
| Rows have exactly three columns: | |
| - `inputs`: single token stream containing byte text, special markers, and flattened SNAC tokens | |
| - `labels`: same length as `inputs`, with `-100` for text/prompt positions | |
| - `attentionmask`: sequence mask, `1` for valid positions | |