Source code for torchaudio.datasets.speechcommands

import os
from pathlib import Path
from typing import Optional, Tuple, Union

import torchaudio
from torch import Tensor
from torch.hub import download_url_to_file
from import Dataset
from torchaudio.datasets.utils import extract_archive

FOLDER_IN_ARCHIVE = "SpeechCommands"
URL = "speech_commands_v0.02"
HASH_DIVIDER = "_nohash_"
EXCEPT_FOLDER = "_background_noise_"
    "": "743935421bb51cccdb6bdd152e04c5c70274e935c82119ad7faeec31780d811d",  # noqa: E501
    "": "af14739ee7dc311471de98f5f9d2c9191b18aedfe957f4a6ff791c709868ff58",  # noqa: E501

def _load_list(root, *filenames):
    output = []
    for filename in filenames:
        filepath = os.path.join(root, filename)
        with open(filepath) as fileobj:
            output += [os.path.normpath(os.path.join(root, line.strip())) for line in fileobj]
    return output

def load_speechcommands_item(filepath: str, path: str) -> Tuple[Tensor, int, str, str, int]:
    relpath = os.path.relpath(filepath, path)
    label, filename = os.path.split(relpath)
    # Besides the officially supported split method for datasets defined by "validation_list.txt"
    # and "testing_list.txt" over "speech_commands_v0.0x.tar.gz" archives, an alternative split
    # method referred to in paragraph 2-3 of Section 7.1, references 13 and 14 of the original
    # paper, and the checksums file from the tensorflow_datasets package [1] is also supported.
    # Some filenames in those "speech_commands_test_set_v0.0x.tar.gz" archives have the form
    # "xxx.wav.wav", so file extensions twice needs to be stripped twice.
    # [1]
    speaker, _ = os.path.splitext(filename)
    speaker, _ = os.path.splitext(speaker)

    speaker_id, utterance_number = speaker.split(HASH_DIVIDER)
    utterance_number = int(utterance_number)

    # Load audio
    waveform, sample_rate = torchaudio.load(filepath)
    return waveform, sample_rate, label, speaker_id, utterance_number

[docs]class SPEECHCOMMANDS(Dataset): """Create a Dataset for *Speech Commands* [:footcite:`speechcommandsv2`]. Args: root (str or Path): Path to the directory where the dataset is found or downloaded. url (str, optional): The URL to download the dataset from, or the type of the dataset to dowload. Allowed type values are ``"speech_commands_v0.01"`` and ``"speech_commands_v0.02"`` (default: ``"speech_commands_v0.02"``) folder_in_archive (str, optional): The top-level directory of the dataset. (default: ``"SpeechCommands"``) download (bool, optional): Whether to download the dataset if it is not found at root path. (default: ``False``). subset (str or None, optional): Select a subset of the dataset [None, "training", "validation", "testing"]. None means the whole dataset. "validation" and "testing" are defined in "validation_list.txt" and "testing_list.txt", respectively, and "training" is the rest. Details for the files "validation_list.txt" and "testing_list.txt" are explained in the README of the dataset and in the introduction of Section 7 of the original paper and its reference 12. The original paper can be found `here <>`_. (Default: ``None``) """ def __init__( self, root: Union[str, Path], url: str = URL, folder_in_archive: str = FOLDER_IN_ARCHIVE, download: bool = False, subset: Optional[str] = None, ) -> None: assert subset is None or subset in ["training", "validation", "testing"], ( "When `subset` not None, it must take a value from " + "{'training', 'validation', 'testing'}." ) if url in [ "speech_commands_v0.01", "speech_commands_v0.02", ]: base_url = "" ext_archive = ".tar.gz" url = os.path.join(base_url, url + ext_archive) # Get string representation of 'root' in case Path object is passed root = os.fspath(root) basename = os.path.basename(url) archive = os.path.join(root, basename) basename = basename.rsplit(".", 2)[0] folder_in_archive = os.path.join(folder_in_archive, basename) self._path = os.path.join(root, folder_in_archive) if download: if not os.path.isdir(self._path): if not os.path.isfile(archive): checksum = _CHECKSUMS.get(url, None) download_url_to_file(url, archive, hash_prefix=checksum) extract_archive(archive, self._path) else: if not os.path.exists(self._path): raise RuntimeError( f"The path {self._path} doesn't exist. " "Please check the ``root`` path or set `download=True` to download it" ) if subset == "validation": self._walker = _load_list(self._path, "validation_list.txt") elif subset == "testing": self._walker = _load_list(self._path, "testing_list.txt") elif subset == "training": excludes = set(_load_list(self._path, "validation_list.txt", "testing_list.txt")) walker = sorted(str(p) for p in Path(self._path).glob("*/*.wav")) self._walker = [ w for w in walker if HASH_DIVIDER in w and EXCEPT_FOLDER not in w and os.path.normpath(w) not in excludes ] else: walker = sorted(str(p) for p in Path(self._path).glob("*/*.wav")) self._walker = [w for w in walker if HASH_DIVIDER in w and EXCEPT_FOLDER not in w]
[docs] def __getitem__(self, n: int) -> Tuple[Tensor, int, str, str, int]: """Load the n-th sample from the dataset. Args: n (int): The index of the sample to be loaded Returns: (Tensor, int, str, str, int): ``(waveform, sample_rate, label, speaker_id, utterance_number)`` """ fileid = self._walker[n] return load_speechcommands_item(fileid, self._path)
def __len__(self) -> int: return len(self._walker)


Access comprehensive developer documentation for PyTorch

View Docs


Get in-depth tutorials for beginners and advanced developers

View Tutorials


Find development resources and get your questions answered

View Resources