torchaudio 2.9.0__cp314-cp314-macosx_11_0_arm64.whl

torchaudio/datasets/utils.py

@@ -0,0 +1,54 @@
+import logging
+import os
+import tarfile
+import zipfile
+from typing import Any, List, Optional  # noqa: F401
+
+import torchaudio
+
+_LG = logging.getLogger(__name__)
+
+
+def _extract_tar(from_path: str, to_path: Optional[str] = None, overwrite: bool = False) -> List[str]:
+    if to_path is None:
+        to_path = os.path.dirname(from_path)
+    with tarfile.open(from_path, "r") as tar:
+        files = []
+        for file_ in tar:  # type: Any
+            file_path = os.path.join(to_path, file_.name)
+            if file_.isfile():
+                files.append(file_path)
+                if os.path.exists(file_path):
+                    _LG.info("%s already extracted.", file_path)
+                    if not overwrite:
+                        continue
+            tar.extract(file_, to_path)
+        return files
+
+
+def _extract_zip(from_path: str, to_path: Optional[str] = None, overwrite: bool = False) -> List[str]:
+    if to_path is None:
+        to_path = os.path.dirname(from_path)
+
+    with zipfile.ZipFile(from_path, "r") as zfile:
+        files = zfile.namelist()
+        for file_ in files:
+            file_path = os.path.join(to_path, file_)
+            if os.path.exists(file_path):
+                _LG.info("%s already extracted.", file_path)
+                if not overwrite:
+                    continue
+            zfile.extract(file_, to_path)
+    return files
+
+
+def _load_waveform(
+    root: str,
+    filename: str,
+    exp_sample_rate: int,
+):
+    path = os.path.join(root, filename)
+    waveform, sample_rate = torchaudio.load(path)
+    if exp_sample_rate != sample_rate:
+        raise ValueError(f"sample rate should be {exp_sample_rate}, but got {sample_rate}")
+    return waveform

torchaudio/datasets/vctk.py

@@ -0,0 +1,143 @@
+import os
+from typing import Tuple
+
+import torchaudio
+from torch import Tensor
+from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
+from torchaudio.datasets.utils import _extract_zip
+
+URL = "https://datashare.is.ed.ac.uk/bitstream/handle/10283/3443/VCTK-Corpus-0.92.zip"
+_CHECKSUMS = {
+    "https://datashare.is.ed.ac.uk/bitstream/handle/10283/3443/VCTK-Corpus-0.92.zip": "f96258be9fdc2cbff6559541aae7ea4f59df3fcaf5cf963aae5ca647357e359c"  # noqa: E501
+}
+
+
+SampleType = Tuple[Tensor, int, str, str, str]
+
+
+class VCTK_092(Dataset):
+    """*VCTK 0.92* :cite:`yamagishi2019vctk` dataset
+
+    Args:
+        root (str): Root directory where the dataset's top level directory is found.
+        mic_id (str, optional): Microphone ID. Either ``"mic1"`` or ``"mic2"``. (default: ``"mic2"``)
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (default: ``False``).
+        url (str, optional): The URL to download the dataset from.
+            (default: ``"https://datashare.is.ed.ac.uk/bitstream/handle/10283/3443/VCTK-Corpus-0.92.zip"``)
+        audio_ext (str, optional): Custom audio extension if dataset is converted to non-default audio format.
+
+    Note:
+        * All the speeches from speaker ``p315`` will be skipped due to the lack of the corresponding text files.
+        * All the speeches from ``p280`` will be skipped for ``mic_id="mic2"`` due to the lack of the audio files.
+        * Some of the speeches from speaker ``p362`` will be skipped due to the lack of  the audio files.
+        * See Also: https://datashare.is.ed.ac.uk/handle/10283/3443
+    """
+
+    def __init__(
+        self,
+        root: str,
+        mic_id: str = "mic2",
+        download: bool = False,
+        url: str = URL,
+        audio_ext=".flac",
+    ):
+        if mic_id not in ["mic1", "mic2"]:
+            raise RuntimeError(f'`mic_id` has to be either "mic1" or "mic2". Found: {mic_id}')
+
+        archive = os.path.join(root, "VCTK-Corpus-0.92.zip")
+
+        self._path = os.path.join(root, "VCTK-Corpus-0.92")
+        self._txt_dir = os.path.join(self._path, "txt")
+        self._audio_dir = os.path.join(self._path, "wav48_silence_trimmed")
+        self._mic_id = mic_id
+        self._audio_ext = audio_ext
+
+        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_zip(archive, self._path)
+
+        if not os.path.isdir(self._path):
+            raise RuntimeError("Dataset not found. Please use `download=True` to download it.")
+
+        # Extracting speaker IDs from the folder structure
+        self._speaker_ids = sorted(os.listdir(self._txt_dir))
+        self._sample_ids = []
+
+        """
+        Due to some insufficient data complexity in the 0.92 version of this dataset,
+        we start traversing the audio folder structure in accordance with the text folder.
+        As some of the audio files are missing of either ``mic_1`` or ``mic_2`` but the
+        text is present for the same, we first check for the existence of the audio file
+        before adding it to the ``sample_ids`` list.
+
+        Once the ``audio_ids`` are loaded into memory we can quickly access the list for
+        different parameters required by the user.
+        """
+        for speaker_id in self._speaker_ids:
+            if speaker_id == "p280" and mic_id == "mic2":
+                continue
+            utterance_dir = os.path.join(self._txt_dir, speaker_id)
+            for utterance_file in sorted(f for f in os.listdir(utterance_dir) if f.endswith(".txt")):
+                utterance_id = os.path.splitext(utterance_file)[0]
+                audio_path_mic = os.path.join(
+                    self._audio_dir,
+                    speaker_id,
+                    f"{utterance_id}_{mic_id}{self._audio_ext}",
+                )
+                if speaker_id == "p362" and not os.path.isfile(audio_path_mic):
+                    continue
+                self._sample_ids.append(utterance_id.split("_"))
+
+    def _load_text(self, file_path) -> str:
+        with open(file_path) as file_path:
+            return file_path.readlines()[0]
+
+    def _load_audio(self, file_path) -> Tuple[Tensor, int]:
+        return torchaudio.load(file_path)
+
+    def _load_sample(self, speaker_id: str, utterance_id: str, mic_id: str) -> SampleType:
+        transcript_path = os.path.join(self._txt_dir, speaker_id, f"{speaker_id}_{utterance_id}.txt")
+        audio_path = os.path.join(
+            self._audio_dir,
+            speaker_id,
+            f"{speaker_id}_{utterance_id}_{mic_id}{self._audio_ext}",
+        )
+
+        # Reading text
+        transcript = self._load_text(transcript_path)
+
+        # Reading FLAC
+        waveform, sample_rate = self._load_audio(audio_path)
+
+        return (waveform, sample_rate, transcript, speaker_id, utterance_id)
+
+    def __getitem__(self, n: int) -> SampleType:
+        """Load the n-th sample from the dataset.
+
+        Args:
+            n (int): The index of the sample to be loaded
+
+        Returns:
+            Tuple of the following items;
+
+            Tensor:
+                Waveform
+            int:
+                Sample rate
+            str:
+                Transcript
+            str:
+                Speaker ID
+            std:
+                Utterance ID
+        """
+        speaker_id, utterance_id = self._sample_ids[n]
+        return self._load_sample(speaker_id, utterance_id, self._mic_id)
+
+    def __len__(self) -> int:
+        return len(self._sample_ids)

torchaudio/datasets/voxceleb1.py

@@ -0,0 +1,309 @@
+import os
+from pathlib import Path
+from typing import List, Tuple, Union
+
+from torch import Tensor
+from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
+from torchaudio.datasets.utils import _extract_zip, _load_waveform
+
+
+SAMPLE_RATE = 16000
+_ARCHIVE_CONFIGS = {
+    "dev": {
+        "archive_name": "vox1_dev_wav.zip",
+        "urls": [
+            "https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a/vox1_dev_wav_partaa",
+            "https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a/vox1_dev_wav_partab",
+            "https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a/vox1_dev_wav_partac",
+            "https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a/vox1_dev_wav_partad",
+        ],
+        "checksums": [
+            "21ec6ca843659ebc2fdbe04b530baa4f191ad4b0971912672d92c158f32226a0",
+            "311d21e0c8cbf33573a4fce6c80e5a279d80736274b381c394319fc557159a04",
+            "92b64465f2b2a3dc0e4196ae8dd6828cbe9ddd1f089419a11e4cbfe2e1750df0",
+            "00e6190c770b27f27d2a3dd26ee15596b17066b715ac111906861a7d09a211a5",
+        ],
+    },
+    "test": {
+        "archive_name": "vox1_test_wav.zip",
+        "url": "https://thor.robots.ox.ac.uk/~vgg/data/voxceleb/vox1a/vox1_test_wav.zip",
+        "checksum": "8de57f347fe22b2c24526e9f444f689ecf5096fc2a92018cf420ff6b5b15eaea",
+    },
+}
+_IDEN_SPLIT_URL = "https://www.robots.ox.ac.uk/~vgg/data/voxceleb/meta/iden_split.txt"
+_VERI_TEST_URL = "https://www.robots.ox.ac.uk/~vgg/data/voxceleb/meta/veri_test.txt"
+
+
+def _download_extract_wavs(root: str):
+    for archive in ["dev", "test"]:
+        archive_name = _ARCHIVE_CONFIGS[archive]["archive_name"]
+        archive_path = os.path.join(root, archive_name)
+        # The zip file of dev data is splited to 4 chunks.
+        # Download and combine them into one file before extraction.
+        if archive == "dev":
+            urls = _ARCHIVE_CONFIGS[archive]["urls"]
+            checksums = _ARCHIVE_CONFIGS[archive]["checksums"]
+            with open(archive_path, "wb") as f:
+                for url, checksum in zip(urls, checksums):
+                    file_path = os.path.join(root, os.path.basename(url))
+                    download_url_to_file(url, file_path, hash_prefix=checksum)
+                    with open(file_path, "rb") as f_split:
+                        f.write(f_split.read())
+        else:
+            url = _ARCHIVE_CONFIGS[archive]["url"]
+            checksum = _ARCHIVE_CONFIGS[archive]["checksum"]
+            download_url_to_file(url, archive_path, hash_prefix=checksum)
+        _extract_zip(archive_path)
+
+
+def _get_flist(root: str, file_path: str, subset: str) -> List[str]:
+    f_list = []
+    if subset == "train":
+        index = 1
+    elif subset == "dev":
+        index = 2
+    else:
+        index = 3
+    with open(file_path, "r") as f:
+        for line in f:
+            id, path = line.split()
+            if int(id) == index:
+                f_list.append(path)
+    return sorted(f_list)
+
+
+def _get_paired_flist(root: str, veri_test_path: str):
+    f_list = []
+    with open(veri_test_path, "r") as f:
+        for line in f:
+            label, path1, path2 = line.split()
+            f_list.append((label, path1, path2))
+    return f_list
+
+
+def _get_file_id(file_path: str, _ext_audio: str):
+    speaker_id, youtube_id, utterance_id = file_path.split("/")[-3:]
+    utterance_id = utterance_id.replace(_ext_audio, "")
+    file_id = "-".join([speaker_id, youtube_id, utterance_id])
+    return file_id
+
+
+class VoxCeleb1(Dataset):
+    """*VoxCeleb1* :cite:`nagrani2017voxceleb` dataset.
+
+    Args:
+        root (str or Path): Path to the directory where the dataset is found or downloaded.
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (Default: ``False``).
+    """
+
+    _ext_audio = ".wav"
+
+    def __init__(self, root: Union[str, Path], download: bool = False) -> None:
+        # Get string representation of 'root' in case Path object is passed
+        root = os.fspath(root)
+        self._path = os.path.join(root, "wav")
+        if not os.path.isdir(self._path):
+            if not download:
+                raise RuntimeError(
+                    f"Dataset not found at {self._path}. Please set `download=True` to download the dataset."
+                )
+            _download_extract_wavs(root)
+
+    def get_metadata(self, n: int):
+        raise NotImplementedError
+
+    def __getitem__(self, n: int):
+        raise NotImplementedError
+
+    def __len__(self) -> int:
+        raise NotImplementedError
+
+
+class VoxCeleb1Identification(VoxCeleb1):
+    """*VoxCeleb1* :cite:`nagrani2017voxceleb` dataset for speaker identification task.
+
+    Each data sample contains the waveform, sample rate, speaker id, and the file id.
+
+    Args:
+        root (str or Path): Path to the directory where the dataset is found or downloaded.
+        subset (str, optional): Subset of the dataset to use. Options: ["train", "dev", "test"]. (Default: ``"train"``)
+        meta_url (str, optional): The url of meta file that contains the list of subset labels and file paths.
+            The format of each row is ``subset file_path". For example: ``1 id10006/nLEBBc9oIFs/00003.wav``.
+            ``1``, ``2``, ``3`` mean ``train``, ``dev``, and ``test`` subest, respectively.
+            (Default: ``"https://www.robots.ox.ac.uk/~vgg/data/voxceleb/meta/iden_split.txt"``)
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (Default: ``False``).
+
+    Note:
+        The file structure of `VoxCeleb1Identification` dataset is as follows:
+
+        └─ root/
+
+         └─ wav/
+
+         └─ speaker_id folders
+
+        Users who pre-downloaded the ``"vox1_dev_wav.zip"`` and ``"vox1_test_wav.zip"`` files need to move
+        the extracted files into the same ``root`` directory.
+    """
+
+    def __init__(
+        self, root: Union[str, Path], subset: str = "train", meta_url: str = _IDEN_SPLIT_URL, download: bool = False
+    ) -> None:
+        super().__init__(root, download)
+        if subset not in ["train", "dev", "test"]:
+            raise ValueError("`subset` must be one of ['train', 'dev', 'test']")
+        # download the iden_split.txt to get the train, dev, test lists.
+        meta_list_path = os.path.join(root, os.path.basename(meta_url))
+        if not os.path.exists(meta_list_path):
+            download_url_to_file(meta_url, meta_list_path)
+        self._flist = _get_flist(self._path, meta_list_path, subset)
+
+    def get_metadata(self, n: int) -> Tuple[str, int, int, str]:
+        """Get metadata for the n-th sample from the dataset. Returns filepath instead of waveform,
+        but otherwise returns the same fields as :py:func:`__getitem__`.
+
+        Args:
+            n (int): The index of the sample
+
+        Returns:
+            Tuple of the following items;
+
+            str:
+                Path to audio
+            int:
+                Sample rate
+            int:
+                Speaker ID
+            str:
+                File ID
+        """
+        file_path = self._flist[n]
+        file_id = _get_file_id(file_path, self._ext_audio)
+        speaker_id = file_id.split("-")[0]
+        speaker_id = int(speaker_id[3:])
+        return file_path, SAMPLE_RATE, speaker_id, file_id
+
+    def __getitem__(self, n: int) -> Tuple[Tensor, int, int, str]:
+        """Load the n-th sample from the dataset.
+
+        Args:
+            n (int): The index of the sample to be loaded
+
+        Returns:
+            Tuple of the following items;
+
+            Tensor:
+                Waveform
+            int:
+                Sample rate
+            int:
+                Speaker ID
+            str:
+                File ID
+        """
+        metadata = self.get_metadata(n)
+        waveform = _load_waveform(self._path, metadata[0], metadata[1])
+        return (waveform,) + metadata[1:]
+
+    def __len__(self) -> int:
+        return len(self._flist)
+
+
+class VoxCeleb1Verification(VoxCeleb1):
+    """*VoxCeleb1* :cite:`nagrani2017voxceleb` dataset for speaker verification task.
+
+    Each data sample contains a pair of waveforms, sample rate, the label indicating if they are
+    from the same speaker, and the file ids.
+
+    Args:
+        root (str or Path): Path to the directory where the dataset is found or downloaded.
+        meta_url (str, optional): The url of meta file that contains a list of utterance pairs
+            and the corresponding labels. The format of each row is ``label file_path1 file_path2".
+            For example: ``1 id10270/x6uYqmx31kE/00001.wav id10270/8jEAjG6SegY/00008.wav``.
+            ``1`` means the two utterances are from the same speaker, ``0`` means not.
+            (Default: ``"https://www.robots.ox.ac.uk/~vgg/data/voxceleb/meta/veri_test.txt"``)
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (Default: ``False``).
+
+    Note:
+        The file structure of `VoxCeleb1Verification` dataset is as follows:
+
+        └─ root/
+
+         └─ wav/
+
+         └─ speaker_id folders
+
+        Users who pre-downloaded the ``"vox1_dev_wav.zip"`` and ``"vox1_test_wav.zip"`` files need to move
+        the extracted files into the same ``root`` directory.
+    """
+
+    def __init__(self, root: Union[str, Path], meta_url: str = _VERI_TEST_URL, download: bool = False) -> None:
+        super().__init__(root, download)
+        # download the veri_test.txt to get the list of training pairs and labels.
+        meta_list_path = os.path.join(root, os.path.basename(meta_url))
+        if not os.path.exists(meta_list_path):
+            download_url_to_file(meta_url, meta_list_path)
+        self._flist = _get_paired_flist(self._path, meta_list_path)
+
+    def get_metadata(self, n: int) -> Tuple[str, str, int, int, str, str]:
+        """Get metadata for the n-th sample from the dataset. Returns filepaths instead of waveforms,
+        but otherwise returns the same fields as :py:func:`__getitem__`.
+
+        Args:
+            n (int): The index of the sample
+
+        Returns:
+            Tuple of the following items;
+
+            str:
+                Path to audio file of speaker 1
+            str:
+                Path to audio file of speaker 2
+            int:
+                Sample rate
+            int:
+                Label
+            str:
+                File ID of speaker 1
+            str:
+                File ID of speaker 2
+        """
+        label, file_path_spk1, file_path_spk2 = self._flist[n]
+        label = int(label)
+        file_id_spk1 = _get_file_id(file_path_spk1, self._ext_audio)
+        file_id_spk2 = _get_file_id(file_path_spk2, self._ext_audio)
+        return file_path_spk1, file_path_spk2, SAMPLE_RATE, label, file_id_spk1, file_id_spk2
+
+    def __getitem__(self, n: int) -> Tuple[Tensor, Tensor, int, int, str, str]:
+        """Load the n-th sample from the dataset.
+
+        Args:
+            n (int): The index of the sample to be loaded.
+
+        Returns:
+            Tuple of the following items;
+
+            Tensor:
+                Waveform of speaker 1
+            Tensor:
+                Waveform of speaker 2
+            int:
+                Sample rate
+            int:
+                Label
+            str:
+                File ID of speaker 1
+            str:
+                File ID of speaker 2
+        """
+        metadata = self.get_metadata(n)
+        waveform_spk1 = _load_waveform(self._path, metadata[0], metadata[2])
+        waveform_spk2 = _load_waveform(self._path, metadata[1], metadata[2])
+        return (waveform_spk1, waveform_spk2) + metadata[2:]
+
+    def __len__(self) -> int:
+        return len(self._flist)

torchaudio/datasets/yesno.py

@@ -0,0 +1,89 @@
+import os
+from pathlib import Path
+from typing import List, Tuple, Union
+
+import torchaudio
+from torch import Tensor
+from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
+from torchaudio.datasets.utils import _extract_tar
+
+
+_RELEASE_CONFIGS = {
+    "release1": {
+        "folder_in_archive": "waves_yesno",
+        "url": "http://www.openslr.org/resources/1/waves_yesno.tar.gz",
+        "checksum": "c3f49e0cca421f96b75b41640749167b52118f232498667ca7a5f9416aef8e73",
+    }
+}
+
+
+class YESNO(Dataset):
+    """*YesNo* :cite:`YesNo` dataset.
+
+    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.
+            (default: ``"http://www.openslr.org/resources/1/waves_yesno.tar.gz"``)
+        folder_in_archive (str, optional):
+            The top-level directory of the dataset. (default: ``"waves_yesno"``)
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (default: ``False``).
+    """
+
+    def __init__(
+        self,
+        root: Union[str, Path],
+        url: str = _RELEASE_CONFIGS["release1"]["url"],
+        folder_in_archive: str = _RELEASE_CONFIGS["release1"]["folder_in_archive"],
+        download: bool = False,
+    ) -> None:
+
+        self._parse_filesystem(root, url, folder_in_archive, download)
+
+    def _parse_filesystem(self, root: str, url: str, folder_in_archive: str, download: bool) -> None:
+        root = Path(root)
+        archive = os.path.basename(url)
+        archive = root / archive
+
+        self._path = root / folder_in_archive
+        if download:
+            if not os.path.isdir(self._path):
+                if not os.path.isfile(archive):
+                    checksum = _RELEASE_CONFIGS["release1"]["checksum"]
+                    download_url_to_file(url, archive, hash_prefix=checksum)
+                _extract_tar(archive)
+
+        if not os.path.isdir(self._path):
+            raise RuntimeError("Dataset not found. Please use `download=True` to download it.")
+
+        self._walker = sorted(str(p.stem) for p in Path(self._path).glob("*.wav"))
+
+    def _load_item(self, fileid: str, path: str):
+        labels = [int(c) for c in fileid.split("_")]
+        file_audio = os.path.join(path, fileid + ".wav")
+        waveform, sample_rate = torchaudio.load(file_audio)
+        return waveform, sample_rate, labels
+
+    def __getitem__(self, n: int) -> Tuple[Tensor, int, List[int]]:
+        """Load the n-th sample from the dataset.
+
+        Args:
+            n (int): The index of the sample to be loaded
+
+        Returns:
+            Tuple of the following items;
+
+            Tensor:
+                Waveform
+            int:
+                Sample rate
+            List[int]:
+                labels
+        """
+        fileid = self._walker[n]
+        item = self._load_item(fileid, self._path)
+        return item
+
+    def __len__(self) -> int:
+        return len(self._walker)