Copy from previous app2

app.py

@@ -0,0 +1,59 @@
+import pandas as pd
+import gradio as gr
+print(gr.__version__)
+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)
+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()
+
+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
+
+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)

audio_loader.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7478d9de0a639a8a769684847bf697c8899e74825d9822e4d90895b916b2b51
+size 23265

requirements.txt

@@ -0,0 +1 @@
+torchaudio