PyTorch Dataloader Raises soundfile.LibsndfileError

Question

I'm running this code and got stuck on this error for too long. I don't know why it's happening, could you help?

Traceback (most recent call last):   File "C:\Users\Catarina Oliveira\PycharmProjects\SER_challenge\main.py", line 12, in <module>     for audio_batch, label_batch, meta_data_batch, b_index in dataloader:   File "C:\Users\Catarina Oliveira\PycharmProjects\SER_challenge\.venv\lib\site-packages\torch\utils\data\dataloader.py", line 631, in __next__     data = self._next_data()   File "C:\Users\Catarina Oliveira\PycharmProjects\SER_challenge\.venv\lib\site-packages\torch\utils\data\dataloader.py", line 1346, in _next_data     return self._process_data(data)   File "C:\Users\Catarina Oliveira\PycharmProjects\SER_challenge\.venv\lib\site-packages\torch\utils\data\dataloader.py", line 1372, in _process_data     data.reraise()   File "C:\Users\Catarina Oliveira\PycharmProjects\SER_challenge\.venv\lib\site-packages\torch\_utils.py", line 722, in reraise     raise exception soundfile.LibsndfileError: <exception str() failed>

My code:

This is the AudioDataset script:

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

class AudioDataset(Dataset):
    def __init__(self, data_csv, data_path, batch_size):
        self.data_path = data_path
        self.data_csv = pd.read_csv(data_path+data_csv)
        self.batch_size = batch_size

        self.audio_files = [os.path.join(self.data_path + '/Audios', file) for file in self.data_csv.iloc[:, 0]]

    def __len__(self):
        return len(self.data_csv) // self.batch_size

    def __getitem__(self, idx):
        start_idx = idx * self.batch_size
        end_idx = start_idx + self.batch_size

        batch_waveforms = []
        batch_labels = []
        batch_meta_data = []

        for idx in range(start_idx, end_idx):
            audio_file = os.path.join(self.data_path+'/Audios', self.data_csv.iloc[idx, 0])
            waveform, sample_rate = torchaudio.load(audio_file)
            batch_waveforms.append(waveform)

            label = self.data_csv.iloc[idx, 1]
            batch_labels.append(label)

            batch_meta_data.append(self.data_csv.iloc[idx])

        max_length = max(sample.shape[1] for sample in batch_waveforms)
        batch_waveforms = [torch.nn.functional.pad(audio_tensor, (0, max_length - audio_tensor.shape[1])) for
                       audio_tensor in batch_waveforms]

        return torch.stack(batch_waveforms), batch_labels, batch_meta_data, idx

This is the main script. I also would like to write the program without the if name=main but don't know how.

from torch.utils.data import DataLoader
from AudioDataset import AudioDataset
from torchaudio import transforms

if __name__ == '__main__':
    bs = 32
    file_path = ''
    train_dataset = AudioDataset(data_csv='Train_data.csv', data_path=file_path, batch_size=bs)
    dataloader = DataLoader(train_dataset, batch_size=None, shuffle=True, num_workers=16,
                            prefetch_factor=2)  # Setting batch_size=None since the dataset already returns batches

    for audio_batch, label_batch, meta_data_batch, b_index in dataloader:
        transform = transforms.MelSpectrogram(n_fft=400, win_length=400, hop_length=160, n_mels=10)
        mel_specgram = transform(audio_batch)

        # mel_spectrogram = torchaudio.transforms.MelSpectrogram()(audio_batch)
        # features = mel_spectrogram.flatten().numpy()

        # print(b_index)
        # print(audio_batch.shape)
        # print(label_batch)
        # print(meta_data_batch)
        pass