Upload README.md with huggingface_hub

README.md

@@ -1,34 +1,38 @@
 ---
-dataset_info:
-  features:
-  - name: audio
-    dtype:
-      audio:
-        sampling_rate: 24000
-  - name: text
-    dtype: string
-  - name: duration
-    dtype: float32
-  - name: source
-    dtype: string
-  - name: speaker
-    dtype: string
-  - name: language
-    dtype: string
-  splits:
-  - name: train
-    num_bytes: 12234641384.63
-    num_examples: 100842
-  - name: validation
-    num_bytes: 249686503.30400002
-    num_examples: 2058
-  download_size: 12375674149
-  dataset_size: 12484327887.934
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
-  - split: validation
-    path: data/validation-*
+language:
+- fr
+license: cc-by-nc-4.0
+task_categories:
+- text-to-speech
+tags:
+- french
+- tts
+- moshi
+- audio
 ---
+
+# French TTS — FT1 Dataset
+
+Dataset pour le **pré-entraînement (FT1)** dans un pipeline de double fine-tuning TTS français.
+
+## Sources
+| Source | Licence | Style |
+|---|---|---|
+| **Emilia-FR** | CC-BY-NC 4.0 | Spontané, varié (podcasts, débats) |
+| **Blizzard NEB** | CC-BY 4.0 | Audiobook studio, haute qualité |
+
+## Stats
+- Train      : 100842 exemples  (156.8h)
+- Validation : 2058 exemples  (3.2h)
+- Sample rate : 24000 Hz
+- Filtres : durée [1.0s–30.0s], SNR ≥ 20.0 dB (Blizzard uniquement)
+
+## Usage
+```python
+from datasets import load_dataset
+ds = load_dataset("Athirham/french-tts-ft1", split="train", streaming=True)
+```
+
+## Pipeline double fine-tuning
+- **FT1** (ce dataset) → sémantique française, phonologie, diversité prosodique
+- **FT2** (SIWIS ~10h) → accent naturel, timbre propre, pas de son robotique