fix train.py

train.py

@@ -1,3 +1,4 @@
+import librosa
 import torch
 from dataclasses import dataclass
 from typing import Any
@@ -39,9 +40,16 @@ class DataCollatorCTCWithPadding:
 
 def prepare_dataset(batch, processor):
     audio = batch["audio"]
-    batch["input_values"] = processor(
-        audio["array"], sampling_rate=audio["sampling_rate"]
-    ).input_values[0]
+    target_sr = processor.feature_extractor.sampling_rate
+    array = audio["array"]
+    if audio["sampling_rate"] != target_sr:
+        array = librosa.resample(
+            array,
+            orig_sr=audio["sampling_rate"],
+            target_sr=target_sr,
+            res_type="kaiser_fast",
+        )
+    batch["input_values"] = processor(array, sampling_rate=target_sr).input_values[0]
     batch["input_length"] = len(batch["input_values"])
     batch["labels"] = processor(text=batch["normalized_text"]).input_ids
     return batch