Fix iterable dataset bug

dataset.py

@@ -18,12 +18,18 @@ class TTSDataset(datasets.IterableDatasetBuilder):
             }),
         )
 
+    def _split_generators(self, dl_manager):
+        return [
+            datasets.SplitGenerator(name=datasets.Split.TRAIN),
+            datasets.SplitGenerator(name=datasets.Split.VALIDATION),
+            datasets.SplitGenerator(name=datasets.Split.TEST),
+        ]
+    
     def _get_streaming_data(self, split):
         root = Path(self._relative_data_dir or ".")
         wav_dir = root / "wav"
         phon_dir = root / "phonemized"
 
-        # Gather speaker splits as before
         speakers = sorted({d.name for d in wav_dir.iterdir() if d.is_dir()})
         random.seed(_SEED)
         random.shuffle(speakers)
@@ -41,7 +47,6 @@ class TTSDataset(datasets.IterableDatasetBuilder):
         else:
             raise ValueError(f"Unknown split: {split}")
 
-        # Now, for streaming, just yield examples
         for spk in use_speakers:
             for wav_file in (wav_dir / spk).glob("*.wav"):
                 if wav_file.name.startswith("._"):
@@ -56,3 +61,6 @@ class TTSDataset(datasets.IterableDatasetBuilder):
                     "audio": str(wav_file),
                     "phoneme": phonemes,
                 }
+
+    def _generate_examples(self, split=None):
+        yield from self._get_streaming_data(split or "train")