Update app.py

app.py

@@ -14,6 +14,9 @@ OUTPUT_OK = (
  <div class="container">
  <div class="row"><h1 style="text-align: center">The provided samples are</h1></div>
  <div class="row"><h1 class="text-success" style="text-align: center">Same Speakers!!!</h1></div>
+ <div class="row"><h1 class="display-1 text-success" style="text-align: center">similarity score: {:.1f}%</h1></div>
+ <div class="row"><tiny style="text-align: center">(Similarity score must be atleast 80% to be considered as same speaker)</small><div class="row">
+
  </div>
 """
 )
@@ -23,6 +26,8 @@ OUTPUT_FAIL = (
  <div class="container">
  <div class="row"><h1 style="text-align: center">The provided samples are from </h1></div>
  <div class="row"><h1 class="text-danger" style="text-align: center">Different Speakers!!!</h1></div> 
+ <div class="row"><h1 class="display-1 text-danger" style="text-align: center">similarity score: {:.1f}%</h1></div>
+ <div class="row"><tiny style="text-align: center">(Similarity score must be atleast 80% to be considered as same speaker)</small><div class="row">
  </div>
 """
 )
@@ -37,9 +42,22 @@ def compare_samples(path1, path2):
  if not (path1 and path2):
  return '<b style="color:red">ERROR: Please record audio for *both* speakers!</b>'
 
- output = model.verify_speakers(path1,path2,THRESHOLD)
-
- return OUTPUT_OK if output else OUTPUT_FAIL
+ embs1 = model.get_embedding(path1).squeeze()
+ embs2 = model.get_embedding(path2).squeeze()
+ 
+ #Length Normalize
+ X = embs1 / torch.linalg.norm(embs1)
+ Y = embs2 / torch.linalg.norm(embs2)
+ 
+ # Score
+ similarity_score = torch.dot(X, Y) / ((torch.dot(X, X) * torch.dot(Y, Y)) ** 0.5)
+ similarity_score = (similarity_score + 1) / 2
+ 
+ # Decision
+ if similarity_score >= THRESHOLD:
+ return OUTPUT_OK.format(similarity_score * 100)
+ else:
+ return OUTPUT_FAIL.format(similarity_score * 100)
 
 
 inputs = [