Fix dataset viewer: restrict config to raw_audio only

README.md

@@ -8,7 +8,9 @@ size_categories:
   - 1K<n<10K
 configs:
   - config_name: default
-    data_dir: raw_audio
+    data_files:
+      - split: train
+        path: "raw_audio/*"
     default: true
 dataset_info:
   features:
@@ -53,6 +55,16 @@ Vietnamese text-to-speech dataset for fine-tuning VieNeu-TTS models. Contains pa
 - **Duration range**: 3-15 seconds per sample
 - **Total size**: ~733 MB (audio)
 
+### Metadata Files
+
+The following CSV files are included for fine-tuning convenience (pipe-delimited, no headers):
+
+| File | Samples | Format |
+|------|---------|--------|
+| `metadata.csv` | 2,402 | `filename\|text` |
+| `metadata_cleaned.csv` | 1,805 | `filename\|text` |
+| `metadata_encoded.csv` | 1,805 | `filename\|text\|vq_codes` |
+
 ## Data Collection
 
 Dataset was generated using [asr-dataset-generation](https://github.com/bkdev98/asr-dataset-generation).
@@ -72,4 +84,4 @@ Audio-text pairs were collected and filtered with the following criteria:
 from datasets import load_dataset
 
 dataset = load_dataset("quocs/hatrang-voice-4h")
-```
+```