File size: 1,839 Bytes
35ebfd2
fa8a5a0
 
 
35ebfd2
fa8a5a0
 
 
 
 
 
 
 
 
0fed8b2
fa8a5a0
 
eb7474e
 
fa8a5a0
 
070850a
fa8a5a0
 
 
 
 
 
 
 
0fed8b2
fa8a5a0
 
 
0fed8b2
c04325f
 
 
 
0fed8b2
 
c04325f
 
 
 
 
 
fa8a5a0
 
 
0d68ce7
0fed8b2
 
 
 
 
 
fa8a5a0
 
 
 
35ebfd2
 
 
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
import gradio as gr
import torch
import torchaudio
from transformers import AutoFeatureExtractor, ASTForAudioClassification

model_name = "MIT/ast-finetuned-audioset-10-10-0.4593"
model = ASTForAudioClassification.from_pretrained(model_name)
feature_extractor = AutoFeatureExtractor.from_pretrained(model_name)

device = torch.device("cpu")
model.to(device)

def classify_sound(file_path):
    wv, sr = torchaudio.load(file_path)
    original_shape = wv.shape

    # Convert to mono
    if wv.shape[0] > 1:
        wv = wv.mean(dim=0, keepdim=True)

    inputs = feature_extractor(
        wv.squeeze().numpy(), sampling_rate=16000, return_tensors="pt"
    )

    with torch.no_grad():
        logits = model(**inputs).logits

    probs = torch.softmax(logits, dim=-1)[0]
    top5 = torch.topk(probs, k=5)

    top5_labels = [
        (model.config.id2label[idx.item()], round(prob.item(), 4))
        for idx, prob in zip(top5.indices, top5.values)
    ]

    top20 = torch.topk(probs, k=20)
    top20_probs = {
        model.config.id2label[idx.item()]: round(prob.item(), 4)
        for idx, prob in zip(top20.indices, top20.values)
    }

    return (
        dict(top5_labels),             
        str(sr),                      
        str(original_shape),         
        top20_probs                 
    )

demo = gr.Interface(
    fn=classify_sound, 
    inputs=gr.Audio(sources="upload", type="filepath"),
    outputs=[
        gr.Label(label = "Top 5 Pred", num_top_classes=5),
        gr.Textbox(label="Sample Rate"),
        gr.Textbox(label="Waveform Shape"),
        gr.JSON(label="All Class Probabilities")
    ],
    title="Audio Classification with AST",
    description="Upload an audio clip (speech, music, ambient sound, etc.). Model: MIT AST fine-tuned on AudioSet (10 classes).",
    live=False,
)

demo.launch()