speech-test commited on
Commit
420400a
β€’
1 Parent(s): bcff2ae

First iteration

Browse files
Files changed (3) hide show
  1. README.md +2 -2
  2. app.py +82 -0
  3. requirements.txt +2 -0
README.md CHANGED
@@ -1,8 +1,8 @@
1
  ---
2
  title: Unispeech Speaker Verification
3
  emoji: πŸ’»
4
- colorFrom: yellow
5
- colorTo: green
6
  sdk: gradio
7
  app_file: app.py
8
  pinned: false
 
1
  ---
2
  title: Unispeech Speaker Verification
3
  emoji: πŸ’»
4
+ colorFrom: indigo
5
+ colorTo: blue
6
  sdk: gradio
7
  app_file: app.py
8
  pinned: false
app.py ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import torch
2
+ import gradio as gr
3
+ from torchaudio.sox_effects import apply_effects_file
4
+ from transformers import AutoFeatureExtractor, AutoModelForAudioXVector
5
+
6
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
7
+
8
+ OUTPUT = """
9
+ <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/css/bootstrap.min.css" integrity="sha256-YvdLHPgkqJ8DVUxjjnGVlMMJtNimJ6dYkowFFvp4kKs=" crossorigin="anonymous">
10
+ <div class="container">
11
+ <div class="row"><h1 style="text-align: center">The speakers are</h1></div>
12
+ <div class="row"><h1 class="display-1" style="text-align: center">{:.1f}%</h1></div>
13
+ <div class="row"><h1 style="text-align: center">similar</h1></div>
14
+ </div>
15
+ """
16
+
17
+ EFFECTS = [
18
+ ["channels", "1"],
19
+ ["rate", "16000"],
20
+ ["gain", "-3.0"],
21
+ ["silence", "1", "0.1", "0.1%", "-1", "0.1", "0.1%"],
22
+ ]
23
+
24
+ model_name = "anton-l/unispeech-sat-base-plus-sv"
25
+ feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)
26
+ model = AutoModelForAudioXVector.from_pretrained(model_name).to(device)
27
+ cosine_sim = torch.nn.CosineSimilarity(dim=-1)
28
+
29
+
30
+ def similarity_fn(mic_path1, file_path1, mic_path2, file_path2):
31
+ if not ((mic_path1 or file_path1) and (mic_path2 or file_path2)):
32
+ return '<b style="color:red">ERROR: Please record or upload audio for *both* speakers!</b>'
33
+
34
+ wav1, _ = apply_effects_file(mic_path1 if mic_path1 else file_path1, EFFECTS)
35
+ wav2, _ = apply_effects_file(mic_path2 if mic_path2 else file_path2, EFFECTS)
36
+
37
+ input1 = feature_extractor(wav1.squeeze(0), return_tensors="pt").input_values.to(device)
38
+ input2 = feature_extractor(wav2.squeeze(0), return_tensors="pt").input_values.to(device)
39
+
40
+ with torch.no_grad():
41
+ emb1 = model(input1).embeddings
42
+ emb2 = model(input2).embeddings
43
+ emb1 = torch.nn.functional.normalize(emb1, dim=-1).cpu()
44
+ emb2 = torch.nn.functional.normalize(emb2, dim=-1).cpu()
45
+ similarity = cosine_sim(emb1, emb2).numpy()[0]
46
+
47
+ return OUTPUT.format(similarity * 100)
48
+
49
+
50
+ inputs = [
51
+ gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Speaker #1"),
52
+ gr.inputs.Audio(source="upload", type="filepath", optional=True, label="or"),
53
+ gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Speaker #2"),
54
+ gr.inputs.Audio(source="upload", type="filepath", optional=True, label="or"),
55
+ ]
56
+ output = gr.outputs.HTML(label="")
57
+
58
+
59
+ description = (
60
+ "Speaker Verification demo based on "
61
+ "UniSpeech-SAT: Universal Speech Representation Learning with Speaker Aware Pre-Training"
62
+ )
63
+ article = (
64
+ "<p style='text-align: center'>"
65
+ "<a href='https://huggingface.co/microsoft/unispeech-sat-large' target='_blank'>πŸŽ™οΈ Learn more about UniSpeech-SAT</a> | "
66
+ "<a href='https://arxiv.org/abs/2110.05752' target='_blank'>πŸ“š Article on ArXiv</a>"
67
+ "</p>"
68
+ )
69
+
70
+ interface = gr.Interface(
71
+ fn=similarity_fn,
72
+ inputs=inputs,
73
+ outputs=output,
74
+ title="Speaker Verification with UniSpeech-SAT",
75
+ description=description,
76
+ article=article,
77
+ layout="horizontal",
78
+ theme="huggingface",
79
+ allow_flagging=False,
80
+ live=False,
81
+ )
82
+ interface.launch(enable_queue=True)
requirements.txt ADDED
@@ -0,0 +1,2 @@
 
 
 
1
+ git+https://github.com/huggingface/transformers
2
+ torchaudio