Update app.py

app.py

@@ -22,22 +22,23 @@ def read_file(wav):
 
 
 def parse_transcription_with_lm(wav_file):
-    speech = convert_file(wav_file)
-    inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
+    input_values = read_file_and_process(wav_file)
 
     with torch.no_grad():
-        logits = model(**inputs).logits
-    int_result = processor.batch_decode(logits.cpu().numpy())
+        logits = model(**input_values).logits
+    int_result = processor.decode(logits.cpu().numpy())
 
     transcription =  int_result.text
     return transcription
 
 
-def convert_file(wav_file):
+def read_file_and_process(wav_file):
     filename = wav_file.split('.')[0]
     convert(wav_file, filename + "16k.wav")
     speech, _ = sf.read(filename + "16k.wav")
-    return speech
+    inputs = processor(speech, sampling_rate=16_000, return_tensors="pt", padding=True)
+    
+    return inputs
 
 def parse(wav_file, applyLM):
     if applyLM:
@@ -46,12 +47,7 @@ def parse(wav_file, applyLM):
         return parse_transcription(wav_file)
 
 def parse_transcription(wav_file):
-    speech = convert_file(wav_file)
-    
-    
-    #speech = read_file(wav_file)
-    input_values = processor(speech, sampling_rate=16_000, return_tensors="pt").input_values
-
+    input_values = read_file_and_process(wav_file)
     logits = model(input_values).logits
     predicted_ids = torch.argmax(logits, dim=-1)