training/dataset.py

from torch.utils.data import Dataset as TorchDataset
import pandas as pd
import torchaudio
import torch

class ChainsawDataset(TorchDataset):

    def __init__(self):
        self.path="../datasets/freesound/"
        self.ds = pd.read_csv(self.path+"labels.csv")

    def __getitem__(self, index):
        file, label = self.ds.iloc[index]
        x, sr = torchaudio.load(self.path+file)
        x = x.squeeze()
        return {
            'audio': {
                'path': file,
                'array': x,
                'sampling_rate': torch.tensor(sr),
            },
            'label': torch.tensor(label)
        }
    
    def __len__(self):
        return len(self.ds)