| --- |
| license: cc-by-nc-sa-4.0 |
| language: |
| - ar |
| task_categories: |
| - automatic-speech-recognition |
| - audio-classification |
| pretty_name: Clean One-Speaker SADA (Split) |
| size_categories: |
| - 10K<n<100K |
| --- |
| |
| # Clean_One_Speaker_SADA_Split |
|
|
| Train/validation/test version of |
| [Sebssihakim/Clean_One_Speaker_SADA](https://huggingface.co/datasets/Sebssihakim/Clean_One_Speaker_SADA), |
| a cleaned single-speaker subset of the SADA corpus (SDAIA / Saudi Broadcasting Authority). |
|
|
| ## Splits |
|
|
| 80/10/10, stratified by `speaker_dialect`, seed 42: |
| train ~30.8k, validation ~3.9k, test ~3.9k rows. |
|
|
| - The `Maghrebi` dialect (7 rows) was removed — too few samples to stratify. |
| - Splits are segment-level: the same source show/speaker may appear in more than one split. |
| - Rare dialects (Yemeni, Shamali, Janubi) have very few validation/test samples; |
| per-dialect metrics on them are not statistically meaningful. |
|
|
| ## Columns |
|
|
| `audio`, `text`, `cleaned_text`, `speaker_age`, `speaker_gender`, `speaker_dialect` |
|
|
| ## License & attribution |
|
|
| Derived from **SADA** (SDAIA & Saudi Broadcasting Authority), CC BY-NC-SA 4.0. |
| This derivative keeps the same license (non-commercial, share-alike). |
|
|
