Update app.py

app.py

@@ -5,7 +5,6 @@ import torch
 import torchaudio
 
 
-df= pd.read_csv('native_words_subset.csv')
 
 torch._C._jit_override_can_fuse_on_cpu(False)
 torch._C._jit_override_can_fuse_on_gpu(False)
@@ -13,47 +12,14 @@ torch._C._jit_set_texpr_fuser_enabled(False)
 torch._C._jit_set_nvfuser_enabled(False)
 
 loader = torch.jit.load("audio_loader.pt")
-model = torch.jit.load('QuartzNet_thunderspeech_3.pt').eval()
+model = torch.jit.load('QuartzNet15x5Base_En_1.pt').eval()
 
 vocab = model.text_transform.vocab.itos
 vocab[-1] = ''
 
-def convert_probs(probs):
-  ids = probs.argmax(1)[0]
-  s = []
-  if vocab[ids[0]]: s.append(vocab[ids[0]])
-  for i in range(1,len(ids)):
-    if ids[i-1] != ids[i]:
-      new = vocab[ids[i]]
-      if new: s.append(new)
-  #return '.'.join(s)
-  return s
- 
   
 def predict(path):
   audio = loader(path)
-  probs = model(audio, torch.tensor(audio.shape[0] * [audio.shape[-1]], device=audio.device))[0]
-  return convert_probs(probs)
- 
- 
-from difflib import SequenceMatcher
+  return model.predict(audio)
 
-def similar(a, b):
-    return SequenceMatcher(None, a, b).ratio()
-
-def compare(chosen_word, path):
-  etalons = [list(val.split('.')) for val in df.loc[df['replica'] == chosen_word, 'transcription'].values]
-  user = predict(path)
-  coeff = 0.0
-  idx=0
-  for i in range(len(etalons)):
-    new_coeff =  similar(user, etalons[i])
-    if new_coeff > coeff:
-      coeff = new_coeff
-      idx=i
-  return f'The similarity coefficient of your pronunciation and the pronunciation of a native speaker is {coeff}. The closer the coefficient is to 1, the better.' + '\nYour pronunciation: [' + ''.join(user) + ']\nClosest native pronunciation: ['  + ''.join(etalons[idx]) + ']'
-
-
-word_choice = gr.inputs.Dropdown(sorted(list(df['replica'].unique())), label="Choose a word")
-
-gr.Interface(fn=compare, inputs=[word_choice, gr.inputs.Audio(source='microphone', type='filepath', optional=True)], outputs= 'text').launch(debug=True)
+gr.Interface(fn=predict, predict=[gr.inputs.Audio(source='microphone', type='filepath'], outputs= 'text').launch(debug=True)