Update app.py

app.py

@@ -15,6 +15,8 @@ logging.basicConfig(
 logger = logging.getLogger(__name__)
 logger.setLevel(logging.DEBUG)
 
+CACHED_MODEL = AutoModelForCTC.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-russian")
+
 def run(input_file, history, model_size="300M"):
     language = "Russian"
     decoding_type = "LM"
@@ -25,18 +27,13 @@ def run(input_file, history, model_size="300M"):
     history = [] 
 
     
-    model = {"model_id": "jonatasgrosman/wav2vec2-large-xlsr-53-russian"} 
-    has_lm = True
-    
-    model_instance = AutoModelForCTC.from_pretrained(model["model_id"])   
-
     if decoding_type == "LM":
-        processor = Wav2Vec2ProcessorWithLM.from_pretrained(model["model_id"])
-        asr = pipeline("automatic-speech-recognition", model=model_instance, tokenizer=processor.tokenizer, 
+        processor = Wav2Vec2ProcessorWithLM.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-russian")
+        asr = pipeline("automatic-speech-recognition", model=CACHED_MODEL , tokenizer=processor.tokenizer, 
                        feature_extractor=processor.feature_extractor, decoder=processor.decoder)
     else:
-        processor = Wav2Vec2Processor.from_pretrained(model["model_id"])
-        asr = pipeline("automatic-speech-recognition", model=model_instance, tokenizer=processor.tokenizer, 
+        processor = Wav2Vec2Processor.from_pretrained("jonatasgrosman/wav2vec2-large-xlsr-53-russian")
+        asr = pipeline("automatic-speech-recognition", model=CACHED_MODEL , tokenizer=processor.tokenizer, 
                        feature_extractor=processor.feature_extractor, decoder=None)
 
     transcription = asr(input_file.name, chunk_length_s=5, stride_length_s=1)["text"]