app.py

import os
import pandas as pd
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import Dataset, DataLoader
import librosa
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import LabelEncoder

# Custom dataset
class AudioDataset(Dataset):
    def __init__(self, audio_paths, labels):
        self.audio_paths = audio_paths
        self.labels = labels

    def __len__(self):
        return len(self.audio_paths)

    def __getitem__(self, idx):
        audio_path = self.audio_paths[idx]
        
        # Load audio file using librosa
        audio, sr = librosa.load(audio_path, sr=16000, mono=True)
        
        # Extract MFCCs
        mfccs = librosa.feature.mfcc(y=audio, sr=sr, n_mfcc=40)
        
        # Pad or truncate to ensure consistent shape
        if mfccs.shape[1] > 100:
            mfccs = mfccs[:, :100]
        else:
            mfccs = np.pad(mfccs, ((0, 0), (0, 100 - mfccs.shape[1])), mode='constant')
        
        # Convert to torch tensor
        mfccs_tensor = torch.FloatTensor(mfccs).unsqueeze(0)  # Add channel dimension
        
        # Convert label to Long tensor
        label = torch.LongTensor([self.labels[idx]])[0]
        
        return mfccs_tensor, label

# CNN Model
class AudioClassifier(nn.Module):
    def __init__(self, num_classes):
        super(AudioClassifier, self).__init__()
        self.conv1 = nn.Conv2d(1, 32, kernel_size=3, stride=1, padding=1)
        self.conv2 = nn.Conv2d(32, 64, kernel_size=3, stride=1, padding=1)
        self.conv3 = nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1)
        self.pool = nn.MaxPool2d(2, 2)
        self.fc1 = nn.Linear(128 * 5 * 12, 128)
        self.fc2 = nn.Linear(128, num_classes)
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.5)

    def forward(self, x):
        x = self.pool(self.relu(self.conv1(x)))
        x = self.pool(self.relu(self.conv2(x)))
        x = self.pool(self.relu(self.conv3(x)))
        x = x.view(-1, 128 * 5 * 12)
        x = self.relu(self.fc1(x))
        x = self.dropout(x)
        x = self.fc2(x)
        return x
model=AudioClassifier(2)
model.load_state_dict(torch.load('audio_classifier_model_2.pth',map_location=torch.device('cpu')))
model.eval()
k=(np.load(file="label_encoder_classes.npy",allow_pickle=True))
classes=k
print(k)

def classify_audio(audio_file):
    # Load and preprocess the audio
    audio, sr = librosa.load(audio_file, sr=16000, mono=True)
    mfccs = librosa.feature.mfcc(y=audio, sr=sr, n_mfcc=40)
    if mfccs.shape[1] > 100:
        mfccs = mfccs[:, :100]
    else:
        mfccs = np.pad(mfccs, ((0, 0), (0, 100 - mfccs.shape[1])), mode='constant')
    
    # Convert to torch tensor
    mfccs_tensor = torch.FloatTensor(mfccs).unsqueeze(0).unsqueeze(0)
    
    # Make prediction
    with torch.no_grad():
        outputs = model(mfccs_tensor)
        _, predicted = torch.max(outputs, 1)
    
    # Convert prediction to label
    predicted_label = predicted.item()
    
    return classes[predicted_label]
import gradio as gr
iface = gr.Interface(
    fn=classify_audio,
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="💓 Heartbeat",
    description=(
        "Upload an audio file of a baby's heartbeat to check for abnormalities.<br>"
        "Trained on: Bhaskaran, A., & Arora, M. (2022). Indian Institute of Science Fetal Heart Sound Database (IIScFHSDB) (version 1.0). PhysioNet. https://doi.org/10.13026/9vvw-cx05.<br>"
        "Original publication: Amrutha, B; Sidhesh Kumar, J; George, S. & Arora, M. Heart rate estimation and validation algorithm for fetal phonocardiography. Physiological Measurement, 2022."
    )
)

# Launch the app
iface.launch()