Upload README.md with huggingface_hub

README.md

@@ -13,6 +13,11 @@ tags:
 - audio
 - multi-speaker
 pretty_name: Hausa TTS Dataset
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: data/**
 ---
 
 # Hausa TTS Dataset (HausaTTSEmbed)
@@ -52,42 +57,72 @@ Each example contains:
 
 ## Usage
 
-### Basic Loading - Simple and Clean! 🎉
+### Recommended: Download All Files First
 
-All audio files are in the `data/` directory, so they auto-download with the dataset:
+To ensure all audio files are available, download the entire dataset first:
 
 ```python
+from huggingface_hub import snapshot_download
 from datasets import load_dataset, Audio
+import os
+
+# Download entire dataset (parquet + all audio files)
+print("Downloading dataset (~2GB)...")
+local_dir = snapshot_download(
+    "Aybee5/HausaTTSEmbed",
+    repo_type="dataset",
+    local_dir="hausa_tts_data"
+)
+
+# Load from downloaded files
+dataset = load_dataset(
+    "parquet",
+    data_files=f"{local_dir}/data/*.parquet",
+    split="train"
+)
 
-# Load dataset - automatically downloads parquet AND audio files
-dataset = load_dataset("Aybee5/HausaTTSEmbed", split="train")
+# Fix audio paths to absolute paths
+dataset = dataset.map(lambda x: {"audio": os.path.join(local_dir, x["audio"]), **x})
 
-# Cast audio column to enable audio loading
+# Cast to Audio type
 dataset = dataset.cast_column("audio", Audio(sampling_rate=22050))
 
-print(f"Loaded {len(dataset)} samples")
+print(f"✓ Loaded {len(dataset)} samples")
 
-# Access first sample - audio is already downloaded!
+# Access sample
 sample = dataset[0]
 print(f"Text: {sample['text']}")
-print(f"Speaker: {sample['speaker_id']}")
 print(f"Audio shape: {sample['audio']['array'].shape}")
-print(f"Sample rate: {sample['audio']['sampling_rate']}")
 ```
 
-**No manual downloads needed!** Audio files download automatically because they're in the `data/` directory.
+### For Unsloth TTS Training (Complete Code)
 
-### For Unsloth TTS Training
-
-Perfect for Unsloth - just plug and play:
+Use this complete code in your Unsloth/Colab notebook:
 
 ```python
+from huggingface_hub import snapshot_download
 from datasets import load_dataset, Audio
+import os
+
+# Step 1: Download entire dataset
+print("Downloading Hausa TTS dataset...")
+local_dir = snapshot_download(
+    "Aybee5/HausaTTSEmbed",
+    repo_type="dataset",
+    local_dir="/content/hausa_tts"  # Use /content/ for Colab
+)
 
-# Load dataset (auto-downloads everything)
-raw_ds = load_dataset("Aybee5/HausaTTSEmbed", split="train")
+# Step 2: Load from downloaded files
+raw_ds = load_dataset(
+    "parquet",
+    data_files=f"{local_dir}/data/*.parquet",
+    split="train"
+)
+
+# Step 3: Fix audio paths
+raw_ds = raw_ds.map(lambda x: {"audio": os.path.join(local_dir, x["audio"]), **x})
 
-# Speaker handling (Unsloth's exact code)
+# Step 4: Speaker handling (Unsloth's exact code)
 speaker_key = "source"
 if "source" not in raw_ds.column_names and "speaker_id" not in raw_ds.column_names:
     print("Unsloth: No speaker found, adding default source")
@@ -96,22 +131,22 @@ if "source" not in raw_ds.column_names and "speaker_id" not in raw_ds.column_nam
 elif "source" not in raw_ds.column_names and "speaker_id" in raw_ds.column_names:
     speaker_key = "speaker_id"
 
-# Resample to target sample rate
+# Step 5: Resample to target sample rate
 target_sampling_rate = 24000
 raw_ds = raw_ds.cast_column("audio", Audio(sampling_rate=target_sampling_rate))
 
-# Optional: Split into train/validation
+print(f"✓ Dataset ready: {len(raw_ds)} samples")
+
+# Step 6: Optional - Split into train/validation
 split_ds = raw_ds.train_test_split(test_size=0.1, seed=42)
 train_ds = split_ds['train']
 val_ds = split_ds['test']
 
-print(f"Train: {len(train_ds)}, Validation: {len(val_ds)}")
-
-# Ready to train! No FileNotFoundError! 🎉
-processed_ds = raw_ds.map(preprocess_example, ...)
+# Step 7: Continue with your Unsloth training!
+# processed_ds = raw_ds.map(preprocess_example, ...)
 ```
 
-**Works in Colab, Kaggle, and anywhere else!** No special configuration needed.
+This will work without FileNotFoundError! 🎉
 
 ### With Transformers