diff --git a/repos/hifi-gan/meldataset.py b/repos/hifi-gan/meldataset.py
index 87dcba7..61a7a59 100644
--- a/repos/hifi-gan/meldataset.py
+++ b/repos/hifi-gan/meldataset.py
@@ -15,18 +15,33 @@
 except:
     pw = None
 
+from tqdm import tqdm
 def check_files(sampling_rate, segment_size, training_files):
     len_training_files = len(training_files)
     training_files = [x for x in training_files if os.path.exists(x)]
     if (len_training_files - len(training_files)) > 0:
         print(len_training_files - len(training_files), "Files don't exist (and have been removed from training)")
-    
-    len_training_files = len(training_files)
-    training_files = [x for x in training_files if len(load_wav_to_torch(x, target_sr=sampling_rate, return_empty_on_exception=True)[0]) > segment_size]
-    if (len_training_files - len(training_files)) > 0:
-        print(len_training_files - len(training_files), "Files are too short (and have been removed from training)")
+    if not os.path.exists("./bad_files.txt"):
+        len_training_files = len(training_files)
+        for training_file in tqdm(training_files):
+            verify = len(load_wav_to_torch(training_file, target_sr=sampling_rate, return_empty_on_exception=True)[0])
+            if verify < segment_size:
+                training_files.remove(training_file)
+                with open("./bad_files.txt", "a") as shit_file:
+                    shit_file.write(training_file + "\n")
+        if (len_training_files - len(training_files)) > 0:
+            print(len_training_files - len(training_files), "Files are too short (and have been removed from training)")
+    else:
+        with open("./bad_files.txt") as f:
+            bad_files = f.read().splitlines()
+            len_training_files = len(training_files)
+            for training_file in tqdm(training_files):
+                if training_file in bad_files:
+                    training_files.remove(training_file)
+            if (len_training_files - len(training_files)) > 0:
+                print(len_training_files - len(training_files), "Files are too short (and have been removed from training)")
+            
     return training_files
-    
 
 def get_dataset_filelist(a, segment_size, sampling_rate):
     if a.input_wavs_dir is None: