Update app.py

app.py

@@ -7,6 +7,7 @@ from transformers.models.whisper.tokenization_whisper import LANGUAGES
 from transformers.pipelines.audio_utils import ffmpeg_read
 
 model_id = "openai/whisper-large-v2"
+device = "cuda" if torch.cuda.is_available() else "CPU"
 
 
 LANGUANGE_MAP = {
@@ -58,19 +59,9 @@ LANGUANGE_MAP = {
  }
 
 
-processor = WhisperProcessor.from_pretrained(model_id)
-model = WhisperForConditionalGeneration.from_pretrained(model_id)
-model.eval()
-model.to(device)
-
-sampling_rate = processor.feature_extractor.sampling_rate
 
-bos_token_id = processor.tokenizer.all_special_ids[-106]
-decoder_input_ids = torch.tensor([bos_token_id]).to(device)
 
 
-device = "cuda" if torch.cuda.is_available() else "CPU"
-
 model_ckpt = "barto17/language-detection-fine-tuned-on-xlm-roberta-base"
 model = AutoModelForSequenceClassification.from_pretrained(model_ckpt)
 tokenizer = AutoTokenizer.from_pretrained(model_ckpt)
@@ -92,6 +83,16 @@ def process_audio_file(file):
     return audio
 
 def transcribe(Microphone, File_Upload):
+    processor = WhisperProcessor.from_pretrained(model_id)
+    model = WhisperForConditionalGeneration.from_pretrained(model_id)
+    model.eval()
+    model.to(device)
+
+    sampling_rate = processor.feature_extractor.sampling_rate
+    
+    bos_token_id = processor.tokenizer.all_special_ids[-106]
+    decoder_input_ids = torch.tensor([bos_token_id]).to(device)
+
     warn_output = ""
     if (Microphone is not None) and (File_Upload is not None):
         warn_output = "WARNING: You've uploaded an audio file and used the microphone. " \