Update app.py

app.py

@@ -33,10 +33,17 @@ model_path = hf_hub_download(repo_id="kobrasoft/kobraspeech-rnn-cs", filename="k
 with tf.keras.utils.custom_object_scope({'CTCLoss': CTCLoss}):
     model = tf.keras.models.load_model(model_path)
 
+num_to_char_path = hf_hub_download(repo_id="kobrasoft/kobraspeech-rnn-cs", filename="num_to_char.json")
+with open(num_to_char_path, "rb") as f:
+    num_to_char = tf.keras.layers.StringLookup(vocabulary=pkl.load(f), oov_token="", invert=True)
+
+def label_to_string(label):
+    return tf.strings.reduce_join(num_to_char(label)).numpy().decode()
+
 def decode_batch_predictions(pred):
     input_len = np.ones(pred.shape[0]) * pred.shape[1]
     # Use greedy search. For complex tasks, you can use beam search
-    results = keras.backend.ctc_decode(pred, input_length=input_len, greedy=True)[0][0]
+    results = tf.keras.backend.ctc_decode(pred, input_length=input_len, greedy=True)[0][0]
     # Iterate over the results and get back the text
     output_text = []
     for result in results: