Update dune_extraction.py

dune_extraction.py

@@ -35,7 +35,7 @@ DATASET_CONFIG = None
 DATASET_SPLIT = "train"
 
 
-# must contain the same bridge key and the expensive audio column. if your metadata (aka DATASET_SOURCE) is not a subsection of this, you can just copy DATASET_SOURCE here.
+# if your metadata and audio dataset are the same, copy DATASET_SOURCE here, otherwise ensure you have a index column (like `key`) so these two can be joined
 AUDIO_DATASET_SOURCE = ["/home/ubuntu/data"]
 AUDIO_DATASET_CONFIG = None
 AUDIO_DATASET_SPLIT = "train"
@@ -66,7 +66,7 @@ TARGET_SAMPLE_RATE = 22050
 INFERENCE_DTYPE = torch.bfloat16
 
 
-MAX_AUDIO_DURATION_SECONDS = 31.0 # tweak this for your own data
+MAX_AUDIO_DURATION_SECONDS = 30.0 # if your clips are capped at another length, tweak it here.