Update app.py

app.py

@@ -10,12 +10,18 @@ feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
 tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)
 model = SpeechEncoderDecoderModel.from_pretrained(model_name).to(device)
 
+if torch.cuda.is_available():
+    model.half()
+
 def process_audio_file(file):
     data, sr = librosa.load(file)
     if sr != 16000:
         data = librosa.resample(data, sr, 16000)
     print(data.shape)
     input_values = feature_extractor(data, return_tensors="pt").input_values.to(device)
+    
+    if torch.cuda.is_available():
+       input_values.to(torch.float16)
     return input_values
     
 def transcribe(file_mic, file_upload, target_language):