mmlearn.datasets.librispeech module¶

LibriSpeech dataset.

class LibriSpeech(root_dir, split='train-clean-100')[source]¶

Bases: Dataset[Example]

LibriSpeech dataset.

This is a wrapper around torchaudio.datasets.LIBRISPEECH that assumes that the dataset is already downloaded and the top-level directory of the dataset in the root directory is librispeech.

Parameters:

root_dir (str) – Root directory of dataset.
split ({"train-clean-100", "train-clean-360", "train-other-500", "dev-clean", "dev-other", "test-clean", "test-other"}, default="train-clean-100") – Split of the dataset to use.

Raises:

ImportError – If torchaudio is not installed.

Notes

This dataset only returns the audio and transcript from the dataset.

__getitem__(idx)[source]¶

Return an example from the dataset.

Return type:: Example

__len__()[source]¶

Return the length of the dataset.

Return type:: int

pad_or_trim(array, length=480000, *, axis=-1)[source]¶

Pad or trim the audio array to length along the given axis.

Parameters:

array (torch.Tensor) – Audio array.
length (int, default=480000) – Length to pad or trim to. Defaults to 30 seconds at 16 kHz.
axis (int, default=-1) – Axis along which to pad or trim.

Returns:

array – Padded or trimmed audio array.

Return type:

torch.Tensor

References