Hugging Face's logo

Spaces:
sulpha
/
titanet-speaker-verification
Build error

App Files Community
titanet-speaker-verification
File size: 3,889 Bytes 
8ed98a1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9876bf6
 
 
8ed98a1
 
 
 
 
 
 
 
 
9876bf6
 
8ed98a1
 
 
 
c762f30
8ed98a1
 
 
 
 
 
 
 
 
9876bf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8ed98a1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
 
import gradio as gr
import torch
from nemo.collections.asr.models import EncDecSpeakerLabelModel


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

STYLE = """
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/css/bootstrap.min.css" integrity="sha256-YvdLHPgkqJ8DVUxjjnGVlMMJtNimJ6dYkowFFvp4kKs=" crossorigin="anonymous">
"""
OUTPUT_OK = (
    STYLE
    + """
    <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>
"""
)
OUTPUT_FAIL = (
    STYLE
    + """
    <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>
"""
)

THRESHOLD = 0.70

model_name = "nvidia/speakerverification_en_titanet_large"
model = EncDecSpeakerLabelModel.from_pretrained(model_name).to(device)


def compare_samples(path1, path2):
    if not (path1 and path2):
        return '<b style="color:red">ERROR: Please record audio for *both* speakers!</b>'

    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 = [
    gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Speaker #1"),
    gr.inputs.Audio(source="microphone", type="filepath", optional=True, label="Speaker #2"),
]
output = gr.outputs.HTML(label="")


description = (
    "This demonstration will analyze two recordings of speech and ascertain whether they have been spoken by the same individual.\n"
    "You can attempt this exercise using your own voice."
)
article = (
    "<p style='text-align: center'>"
    "<a href='https://huggingface.co/nvidia/speakerverification_en_titanet_large' target='_blank'>🎙️ Learn more about TitaNet model</a> | "
    "<a href='https://arxiv.org/pdf/2110.04410.pdf' target='_blank'>📚 TitaNet paper</a> | "
    "<a href='https://github.com/NVIDIA/NeMo' target='_blank'>🧑‍💻 Repository</a>"
    "</p>"
)
examples = [
    ["data/id10270_5r0dWxy17C8-00001.wav", "data/id10270_5r0dWxy17C8-00002.wav"],
    ["data/id10271_1gtz-CUIygI-00001.wav", "data/id10271_1gtz-CUIygI-00002.wav"],
    ["data/id10270_5r0dWxy17C8-00001.wav", "data/id10271_1gtz-CUIygI-00001.wav"],
    ["data/id10270_5r0dWxy17C8-00002.wav", "data/id10271_1gtz-CUIygI-00002.wav"],
]

interface = gr.Interface(
    fn=compare_samples,
    inputs=inputs,
    outputs=output,
    title="Speaker Verification with TitaNet Embeddings",
    description=description,
    article=article,
    layout="horizontal",
    theme="huggingface",
    allow_flagging=False,
    live=False,
    examples=examples,
)
interface.launch(enable_queue=True)