torchaudio 2.0.2__cp310-cp310-win_amd64.whl → 2.1.1__cp310-cp310-win_amd64.whl

torchaudio/datasets/cmudict.py

@@ -3,78 +3,77 @@ import re
 from pathlib import Path
 from typing import Iterable, List, Tuple, Union
 
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
+
 
 _CHECKSUMS = {
     "http://svn.code.sf.net/p/cmusphinx/code/trunk/cmudict/cmudict-0.7b": "209a8b4cd265013e96f4658632a9878103b0c5abf62b50d4ef3ae1be226b29e4",  # noqa: E501
     "http://svn.code.sf.net/p/cmusphinx/code/trunk/cmudict/cmudict-0.7b.symbols": "408ccaae803641c6d7b626b6299949320c2dbca96b2220fd3fb17887b023b027",  # noqa: E501
 }
-_PUNCTUATIONS = set(
-    [
-        "!EXCLAMATION-POINT",
-        '"CLOSE-QUOTE',
-        '"DOUBLE-QUOTE',
-        '"END-OF-QUOTE',
-        '"END-QUOTE',
-        '"IN-QUOTES',
-        '"QUOTE',
-        '"UNQUOTE',
-        "#HASH-MARK",
-        "#POUND-SIGN",
-        "#SHARP-SIGN",
-        "%PERCENT",
-        "&AMPERSAND",
-        "'END-INNER-QUOTE",
-        "'END-QUOTE",
-        "'INNER-QUOTE",
-        "'QUOTE",
-        "'SINGLE-QUOTE",
-        "(BEGIN-PARENS",
-        "(IN-PARENTHESES",
-        "(LEFT-PAREN",
-        "(OPEN-PARENTHESES",
-        "(PAREN",
-        "(PARENS",
-        "(PARENTHESES",
-        ")CLOSE-PAREN",
-        ")CLOSE-PARENTHESES",
-        ")END-PAREN",
-        ")END-PARENS",
-        ")END-PARENTHESES",
-        ")END-THE-PAREN",
-        ")PAREN",
-        ")PARENS",
-        ")RIGHT-PAREN",
-        ")UN-PARENTHESES",
-        "+PLUS",
-        ",COMMA",
-        "--DASH",
-        "-DASH",
-        "-HYPHEN",
-        "...ELLIPSIS",
-        ".DECIMAL",
-        ".DOT",
-        ".FULL-STOP",
-        ".PERIOD",
-        ".POINT",
-        "/SLASH",
-        ":COLON",
-        ";SEMI-COLON",
-        ";SEMI-COLON(1)",
-        "?QUESTION-MARK",
-        "{BRACE",
-        "{LEFT-BRACE",
-        "{OPEN-BRACE",
-        "}CLOSE-BRACE",
-        "}RIGHT-BRACE",
-    ]
-)
+_PUNCTUATIONS = {
+    "!EXCLAMATION-POINT",
+    '"CLOSE-QUOTE',
+    '"DOUBLE-QUOTE',
+    '"END-OF-QUOTE',
+    '"END-QUOTE',
+    '"IN-QUOTES',
+    '"QUOTE',
+    '"UNQUOTE',
+    "#HASH-MARK",
+    "#POUND-SIGN",
+    "#SHARP-SIGN",
+    "%PERCENT",
+    "&AMPERSAND",
+    "'END-INNER-QUOTE",
+    "'END-QUOTE",
+    "'INNER-QUOTE",
+    "'QUOTE",
+    "'SINGLE-QUOTE",
+    "(BEGIN-PARENS",
+    "(IN-PARENTHESES",
+    "(LEFT-PAREN",
+    "(OPEN-PARENTHESES",
+    "(PAREN",
+    "(PARENS",
+    "(PARENTHESES",
+    ")CLOSE-PAREN",
+    ")CLOSE-PARENTHESES",
+    ")END-PAREN",
+    ")END-PARENS",
+    ")END-PARENTHESES",
+    ")END-THE-PAREN",
+    ")PAREN",
+    ")PARENS",
+    ")RIGHT-PAREN",
+    ")UN-PARENTHESES",
+    "+PLUS",
+    ",COMMA",
+    "--DASH",
+    "-DASH",
+    "-HYPHEN",
+    "...ELLIPSIS",
+    ".DECIMAL",
+    ".DOT",
+    ".FULL-STOP",
+    ".PERIOD",
+    ".POINT",
+    "/SLASH",
+    ":COLON",
+    ";SEMI-COLON",
+    ";SEMI-COLON(1)",
+    "?QUESTION-MARK",
+    "{BRACE",
+    "{LEFT-BRACE",
+    "{OPEN-BRACE",
+    "}CLOSE-BRACE",
+    "}RIGHT-BRACE",
+}
 
 
 def _parse_dictionary(lines: Iterable[str], exclude_punctuations: bool) -> List[str]:
     _alt_re = re.compile(r"\([0-9]+\)")
-    cmudict: List[Tuple[str, List[str]]] = list()
+    cmudict: List[Tuple[str, List[str]]] = []
     for line in lines:
         if not line or line.startswith(";;;"):  # ignore comments
             continue

torchaudio/datasets/dr_vctk.py

@@ -3,8 +3,8 @@ from typing import Dict, Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_zip
 
 

torchaudio/datasets/gtzan.py

@@ -4,8 +4,8 @@ from typing import Optional, Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar
 
 # The following lists prefixed with `filtered_` provide a filtered split

torchaudio/datasets/librilight_limited.py

@@ -4,8 +4,8 @@ from typing import List, Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.librispeech import _get_librispeech_metadata
 from torchaudio.datasets.utils import _extract_tar
 

torchaudio/datasets/librispeech.py

@@ -3,8 +3,8 @@ from pathlib import Path
 from typing import Tuple, Union
 
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar, _load_waveform
 
 URL = "train-clean-100"

torchaudio/datasets/librispeech_biasing.py

@@ -0,0 +1,189 @@
+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_tar, _load_waveform
+
+URL = "train-clean-100"
+FOLDER_IN_ARCHIVE = "LibriSpeech"
+SAMPLE_RATE = 16000
+_DATA_SUBSETS = [
+    "dev-clean",
+    "dev-other",
+    "test-clean",
+    "test-other",
+    "train-clean-100",
+    "train-clean-360",
+    "train-other-500",
+]
+_CHECKSUMS = {
+    "http://www.openslr.org/resources/12/dev-clean.tar.gz": "76f87d090650617fca0cac8f88b9416e0ebf80350acb97b343a85fa903728ab3",  # noqa: E501
+    "http://www.openslr.org/resources/12/dev-other.tar.gz": "12661c48e8c3fe1de2c1caa4c3e135193bfb1811584f11f569dd12645aa84365",  # noqa: E501
+    "http://www.openslr.org/resources/12/test-clean.tar.gz": "39fde525e59672dc6d1551919b1478f724438a95aa55f874b576be21967e6c23",  # noqa: E501
+    "http://www.openslr.org/resources/12/test-other.tar.gz": "d09c181bba5cf717b3dee7d4d592af11a3ee3a09e08ae025c5506f6ebe961c29",  # noqa: E501
+    "http://www.openslr.org/resources/12/train-clean-100.tar.gz": "d4ddd1d5a6ab303066f14971d768ee43278a5f2a0aa43dc716b0e64ecbbbf6e2",  # noqa: E501
+    "http://www.openslr.org/resources/12/train-clean-360.tar.gz": "146a56496217e96c14334a160df97fffedd6e0a04e66b9c5af0d40be3c792ecf",  # noqa: E501
+    "http://www.openslr.org/resources/12/train-other-500.tar.gz": "ddb22f27f96ec163645d53215559df6aa36515f26e01dd70798188350adcb6d2",  # noqa: E501
+}
+
+
+def _download_librispeech(root, url):
+    base_url = "http://www.openslr.org/resources/12/"
+    ext_archive = ".tar.gz"
+
+    filename = url + ext_archive
+    archive = os.path.join(root, filename)
+    download_url = os.path.join(base_url, filename)
+    if not os.path.isfile(archive):
+        checksum = _CHECKSUMS.get(download_url, None)
+        download_url_to_file(download_url, archive, hash_prefix=checksum)
+    _extract_tar(archive)
+
+
+def _get_librispeech_metadata(
+    fileid: str, root: str, folder: str, ext_audio: str, ext_txt: str, blist: List[str]
+) -> Tuple[str, int, str, int, int, int]:
+    blist = blist or []
+    speaker_id, chapter_id, utterance_id = fileid.split("-")
+
+    # Get audio path and sample rate
+    fileid_audio = f"{speaker_id}-{chapter_id}-{utterance_id}"
+    filepath = os.path.join(folder, speaker_id, chapter_id, f"{fileid_audio}{ext_audio}")
+
+    # Load text
+    file_text = f"{speaker_id}-{chapter_id}{ext_txt}"
+    file_text = os.path.join(root, folder, speaker_id, chapter_id, file_text)
+    uttblist = []
+    with open(file_text) as ft:
+        for line in ft:
+            fileid_text, transcript = line.strip().split(" ", 1)
+            if fileid_audio == fileid_text:
+                # get utterance biasing list
+                for word in transcript.split():
+                    if word in blist and word not in uttblist:
+                        uttblist.append(word)
+                break
+        else:
+            # Translation not found
+            raise FileNotFoundError(f"Translation not found for {fileid_audio}")
+
+    return (
+        filepath,
+        SAMPLE_RATE,
+        transcript,
+        int(speaker_id),
+        int(chapter_id),
+        int(utterance_id),
+        uttblist,
+    )
+
+
+class LibriSpeechBiasing(Dataset):
+    """*LibriSpeech* :cite:`7178964` dataset with prefix-tree construction and biasing support.
+
+    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 ``"dev-clean"``, ``"dev-other"``, ``"test-clean"``,
+            ``"test-other"``, ``"train-clean-100"``, ``"train-clean-360"`` and
+            ``"train-other-500"``. (default: ``"train-clean-100"``)
+        folder_in_archive (str, optional):
+            The top-level directory of the dataset. (default: ``"LibriSpeech"``)
+        download (bool, optional):
+            Whether to download the dataset if it is not found at root path. (default: ``False``).
+        blist (list, optional):
+            The list of biasing words (default: ``[]``).
+    """
+
+    _ext_txt = ".trans.txt"
+    _ext_audio = ".flac"
+
+    def __init__(
+        self,
+        root: Union[str, Path],
+        url: str = URL,
+        folder_in_archive: str = FOLDER_IN_ARCHIVE,
+        download: bool = False,
+        blist: List[str] = None,
+    ) -> None:
+        self._url = url
+        if url not in _DATA_SUBSETS:
+            raise ValueError(f"Invalid url '{url}' given; please provide one of {_DATA_SUBSETS}.")
+
+        root = os.fspath(root)
+        self._archive = os.path.join(root, folder_in_archive)
+        self._path = os.path.join(root, folder_in_archive, url)
+
+        if not os.path.isdir(self._path):
+            if download:
+                _download_librispeech(root, url)
+            else:
+                raise RuntimeError(
+                    f"Dataset not found at {self._path}. Please set `download=True` to download the dataset."
+                )
+
+        self._walker = sorted(str(p.stem) for p in Path(self._path).glob("*/*/*" + self._ext_audio))
+        self.blist = blist
+
+    def get_metadata(self, n: int) -> Tuple[Tensor, int, str, int, int, int]:
+        """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 to be loaded
+
+        Returns:
+            Tuple of the following items;
+
+            str:
+                Path to audio
+            int:
+                Sample rate
+            str:
+                Transcript
+            int:
+                Speaker ID
+            int:
+                Chapter ID
+            int:
+                Utterance ID
+            list:
+                List of biasing words in the utterance
+        """
+        fileid = self._walker[n]
+        return _get_librispeech_metadata(fileid, self._archive, self._url, self._ext_audio, self._ext_txt, self.blist)
+
+    def __getitem__(self, n: int) -> Tuple[Tensor, int, str, int, int, 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
+            str:
+                Transcript
+            int:
+                Speaker ID
+            int:
+                Chapter ID
+            int:
+                Utterance ID
+            list:
+                List of biasing words in the utterance
+        """
+        metadata = self.get_metadata(n)
+        waveform = _load_waveform(self._archive, metadata[0], metadata[1])
+        return (waveform,) + metadata[1:]
+
+    def __len__(self) -> int:
+        return len(self._walker)

torchaudio/datasets/libritts.py

@@ -4,8 +4,8 @@ from typing import Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar
 
 URL = "train-clean-100"

torchaudio/datasets/ljspeech.py

@@ -5,8 +5,8 @@ from typing import Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar
 
 

torchaudio/datasets/musdb_hq.py

@@ -4,8 +4,8 @@ from typing import List, Optional, Tuple, Union
 
 import torch
 import torchaudio
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_zip
 
 _URL = "https://zenodo.org/record/3338373/files/musdb18hq.zip"

torchaudio/datasets/quesst14.py

@@ -4,8 +4,8 @@ from pathlib import Path
 from typing import Optional, Tuple, Union
 
 import torch
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar, _load_waveform
 
 

torchaudio/datasets/speechcommands.py

@@ -3,8 +3,8 @@ from pathlib import Path
 from typing import Optional, Tuple, Union
 
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar, _load_waveform
 
 FOLDER_IN_ARCHIVE = "SpeechCommands"

torchaudio/datasets/tedlium.py

@@ -4,8 +4,8 @@ from typing import Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar
 
 

torchaudio/datasets/vctk.py

@@ -3,8 +3,8 @@ from typing import Tuple
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 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"

torchaudio/datasets/voxceleb1.py

@@ -3,8 +3,8 @@ from pathlib import Path
 from typing import List, Tuple, Union
 
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_zip, _load_waveform
 
 

torchaudio/datasets/yesno.py

@@ -4,8 +4,8 @@ from typing import List, Tuple, Union
 
 import torchaudio
 from torch import Tensor
-from torch.hub import download_url_to_file
 from torch.utils.data import Dataset
+from torchaudio._internal import download_url_to_file
 from torchaudio.datasets.utils import _extract_tar
 
 

torchaudio/functional/__init__.py

@@ -1,3 +1,4 @@
+from ._alignment import forced_align, merge_tokens, TokenSpan
 from .filtering import (
     allpass_biquad,
     band_biquad,
@@ -28,7 +29,6 @@ from .functional import (
     apply_beamforming,
     apply_codec,
     compute_deltas,
-    compute_kaldi_pitch,
     convolve,
     create_dct,
     DB_to_amplitude,
@@ -36,6 +36,7 @@ from .functional import (
     detect_pitch_frequency,
     edit_distance,
     fftconvolve,
+    frechet_distance,
     griffinlim,
     inverse_spectrogram,
     linear_fbanks,
@@ -64,7 +65,6 @@ from .functional import (
 __all__ = [
     "amplitude_to_DB",
     "compute_deltas",
-    "compute_kaldi_pitch",
     "create_dct",
     "melscale_fbanks",
     "linear_fbanks",
@@ -94,6 +94,9 @@ __all__ = [
     "equalizer_biquad",
     "filtfilt",
     "flanger",
+    "forced_align",
+    "merge_tokens",
+    "TokenSpan",
     "gain",
     "highpass_biquad",
     "lfilter",
@@ -120,4 +123,5 @@ __all__ = [
     "speed",
     "preemphasis",
     "deemphasis",
+    "frechet_distance",
 ]

torchaudio/functional/_alignment.py

@@ -0,0 +1,128 @@
+from dataclasses import dataclass
+from typing import List, Optional, Tuple
+
+import torch
+from torch import Tensor
+from torchaudio._extension import fail_if_no_align
+
+__all__ = []
+
+
+@fail_if_no_align
+def forced_align(
+    log_probs: Tensor,
+    targets: Tensor,
+    input_lengths: Optional[Tensor] = None,
+    target_lengths: Optional[Tensor] = None,
+    blank: int = 0,
+) -> Tuple[Tensor, Tensor]:
+    r"""Align a CTC label sequence to an emission.
+
+    .. devices:: CPU CUDA
+
+    .. properties:: TorchScript
+
+    Args:
+        log_probs (Tensor): log probability of CTC emission output.
+            Tensor of shape `(B, T, C)`. where `B` is the batch size, `T` is the input length,
+            `C` is the number of characters in alphabet including blank.
+        targets (Tensor): Target sequence. Tensor of shape `(B, L)`,
+            where `L` is the target length.
+        input_lengths (Tensor or None, optional):
+            Lengths of the inputs (max value must each be <= `T`). 1-D Tensor of shape `(B,)`.
+        target_lengths (Tensor or None, optional):
+            Lengths of the targets. 1-D Tensor of shape `(B,)`.
+        blank_id (int, optional): The index of blank symbol in CTC emission. (Default: 0)
+
+    Returns:
+        Tuple(Tensor, Tensor):
+            Tensor: Label for each time step in the alignment path computed using forced alignment.
+
+            Tensor: Log probability scores of the labels for each time step.
+
+    Note:
+        The sequence length of `log_probs` must satisfy:
+
+
+        .. math::
+            L_{\text{log\_probs}} \ge L_{\text{label}} + N_{\text{repeat}}
+
+        where :math:`N_{\text{repeat}}` is the number of consecutively repeated tokens.
+        For example, in str `"aabbc"`, the number of repeats are `2`.
+
+    Note:
+        The current version only supports ``batch_size==1``.
+    """
+    if blank in targets:
+        raise ValueError(f"targets Tensor shouldn't contain blank index. Found {targets}.")
+    if torch.max(targets) >= log_probs.shape[-1]:
+        raise ValueError("targets values must be less than the CTC dimension")
+
+    if input_lengths is None:
+        batch_size, length = log_probs.size(0), log_probs.size(1)
+        input_lengths = torch.full((batch_size,), length, dtype=torch.int64, device=log_probs.device)
+    if target_lengths is None:
+        batch_size, length = targets.size(0), targets.size(1)
+        target_lengths = torch.full((batch_size,), length, dtype=torch.int64, device=targets.device)
+
+    # For TorchScript compatibility
+    assert input_lengths is not None
+    assert target_lengths is not None
+
+    paths, scores = torch.ops.torchaudio.forced_align(log_probs, targets, input_lengths, target_lengths, blank)
+    return paths, scores
+
+
+@dataclass
+class TokenSpan:
+    """TokenSpan()
+    Token with time stamps and score. Returned by :py:func:`merge_tokens`.
+    """
+
+    token: int
+    """The token"""
+    start: int
+    """The start time (inclusive) in emission time axis."""
+    end: int
+    """The end time (exclusive) in emission time axis."""
+    score: float
+    """The score of the this token."""
+
+    def __len__(self) -> int:
+        """Returns the time span"""
+        return self.end - self.start
+
+
+def merge_tokens(tokens: Tensor, scores: Tensor, blank: int = 0) -> List[TokenSpan]:
+    """Removes repeated tokens and blank tokens from the given CTC token sequence.
+
+    Args:
+        tokens (Tensor): Alignment tokens (unbatched) returned from :py:func:`forced_align`.
+            Shape: `(time, )`.
+        scores (Tensor): Alignment scores (unbatched) returned from :py:func:`forced_align`.
+            Shape: `(time, )`. When computing the token-size score, the given score is averaged
+            across the corresponding time span.
+
+    Returns:
+        list of TokenSpan
+
+    Example:
+        >>> aligned_tokens, scores = forced_align(emission, targets, input_lengths, target_lengths)
+        >>> token_spans = merge_tokens(aligned_tokens[0], scores[0])
+    """
+    if tokens.ndim != 1 or scores.ndim != 1:
+        raise ValueError("`tokens` and `scores` must be 1D Tensor.")
+    if len(tokens) != len(scores):
+        raise ValueError("`tokens` and `scores` must be the same length.")
+
+    diff = torch.diff(
+        tokens, prepend=torch.tensor([-1], device=tokens.device), append=torch.tensor([-1], device=tokens.device)
+    )
+    changes_wo_blank = torch.nonzero((diff != 0)).squeeze().tolist()
+    tokens = tokens.tolist()
+    spans = [
+        TokenSpan(token=token, start=start, end=end, score=scores[start:end].mean().item())
+        for start, end in zip(changes_wo_blank[:-1], changes_wo_blank[1:])
+        if (token := tokens[start]) != blank
+    ]
+    return spans