torchaudio 2.7.1__cp312-cp312-win_amd64.whl → 2.8.0__cp312-cp312-win_amd64.whl

torchaudio/io/__init__.py

@@ -1,7 +1,14 @@
-from torio.io import CodecConfig, StreamingMediaDecoder as StreamReader, StreamingMediaEncoder as StreamWriter
+from torio.io import CodecConfig as _CodecConfig, StreamingMediaDecoder as _StreamReader, StreamingMediaEncoder as _StreamWriter
+from torchaudio._internal.module_utils import dropping_class_io_support, dropping_class_support, dropping_io_support
 
-from ._effector import AudioEffector
-from ._playback import play_audio
+from ._effector import AudioEffector as _AudioEffector
+from ._playback import play_audio as _play_audio
+
+CodecConfig = dropping_class_io_support(_CodecConfig)
+StreamReader = dropping_class_io_support(_StreamReader)
+StreamWriter = dropping_class_io_support(_StreamWriter)
+AudioEffector = dropping_class_support(_AudioEffector)
+play_audio = dropping_io_support(_play_audio)
 
 
 __all__ = [

torchaudio/kaldi_io.py

@@ -6,6 +6,7 @@ from typing import Any, Callable, Iterable, Tuple
 import torch
 from torch import Tensor
 from torchaudio._internal import module_utils as _mod_utils
+from torchaudio._internal.module_utils import dropping_support
 
 if _mod_utils.is_module_available("numpy"):
     import numpy as np
@@ -41,6 +42,7 @@ def _convert_method_output_to_tensor(
         yield key, torch.from_numpy(np_arr)
 
 
+@dropping_support
 @_mod_utils.requires_module("kaldi_io", "numpy")
 def read_vec_int_ark(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     r"""Create generator of (key,vector<int>) tuples, which reads from the ark file/stream.
@@ -64,6 +66,7 @@ def read_vec_int_ark(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     return _convert_method_output_to_tensor(file_or_fd, kaldi_io.read_vec_int_ark, convert_contiguous=True)
 
 
+@dropping_support
 @_mod_utils.requires_module("kaldi_io", "numpy")
 def read_vec_flt_scp(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     r"""Create generator of (key,vector<float32/float64>) tuples, read according to Kaldi scp.
@@ -84,6 +87,7 @@ def read_vec_flt_scp(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     return _convert_method_output_to_tensor(file_or_fd, kaldi_io.read_vec_flt_scp)
 
 
+@dropping_support
 @_mod_utils.requires_module("kaldi_io", "numpy")
 def read_vec_flt_ark(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     r"""Create generator of (key,vector<float32/float64>) tuples, which reads from the ark file/stream.
@@ -104,6 +108,7 @@ def read_vec_flt_ark(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     return _convert_method_output_to_tensor(file_or_fd, kaldi_io.read_vec_flt_ark)
 
 
+@dropping_support
 @_mod_utils.requires_module("kaldi_io", "numpy")
 def read_mat_scp(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     r"""Create generator of (key,matrix<float32/float64>) tuples, read according to Kaldi scp.
@@ -124,6 +129,7 @@ def read_mat_scp(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     return _convert_method_output_to_tensor(file_or_fd, kaldi_io.read_mat_scp)
 
 
+@dropping_support
 @_mod_utils.requires_module("kaldi_io", "numpy")
 def read_mat_ark(file_or_fd: Any) -> Iterable[Tuple[str, Tensor]]:
     r"""Create generator of (key,matrix<float32/float64>) tuples, which reads from the ark file/stream.

torchaudio/models/decoder/__init__.py

@@ -1,3 +1,5 @@
+from torchaudio._internal.module_utils import dropping_support, dropping_class_support
+import inspect
 _CTC_DECODERS = [
     "CTCHypothesis",
     "CTCDecoder",
@@ -33,7 +35,11 @@ def __getattr__(name: str):
                 "To use CUCTC decoder, please set BUILD_CUDA_CTC_DECODER=1 when building from source."
             ) from err
 
-        item = getattr(_cuda_ctc_decoder, name)
+        orig_item = getattr(_cuda_ctc_decoder, name)
+        if inspect.isclass(orig_item):
+            item = dropping_class_support(orig_item)
+        else:
+            item = dropping_support(orig_item)
         globals()[name] = item
         return item
     raise AttributeError(f"module {__name__} has no attribute {name}")

torchaudio/pipelines/_tts/utils.py

@@ -161,6 +161,7 @@ def _load_phonemizer(file, dl_kwargs):
         raise RuntimeError("DeepPhonemizer is not installed. Please install it.")
 
     from dp.phonemizer import Phonemizer
+    from dp.preprocessing.text import Preprocessor, LanguageTokenizer, SequenceTokenizer
 
     # By default, dp issues DEBUG level log.
     logger = logging.getLogger("dp")
@@ -174,7 +175,8 @@ def _load_phonemizer(file, dl_kwargs):
         if not os.path.exists(path):
             dl_kwargs = {} if dl_kwargs is None else dl_kwargs
             download_url_to_file(url, path, **dl_kwargs)
-        return Phonemizer.from_checkpoint(path)
+        with torch.serialization.safe_globals([Preprocessor, LanguageTokenizer, SequenceTokenizer]):
+            return Phonemizer.from_checkpoint(path)
     finally:
         logger.setLevel(orig_level)
 

torchaudio/prototype/datasets/musan.py

@@ -4,12 +4,13 @@ from typing import Tuple, Union
 import torch
 from torch.utils.data import Dataset
 from torchaudio.datasets.utils import _load_waveform
+from torchaudio._internal.module_utils import dropping_support, dropping_class_support
 
 
 _SUBSETS = ["music", "noise", "speech"]
 _SAMPLE_RATE = 16_000
 
-
+@dropping_class_support
 class Musan(Dataset):
     r"""*MUSAN* :cite:`musan2015` dataset.
 

torchaudio/prototype/functional/_dsp.py

@@ -4,8 +4,10 @@ from typing import List, Optional, Union
 import torch
 
 from torchaudio.functional import fftconvolve
+from torchaudio._internal.module_utils import dropping_support
 
 
+@dropping_support
 def oscillator_bank(
     frequencies: torch.Tensor,
     amplitudes: torch.Tensor,
@@ -81,6 +83,7 @@ def oscillator_bank(
     return waveform
 
 
+@dropping_support
 def adsr_envelope(
     num_frames: int,
     *,
@@ -182,6 +185,7 @@ def adsr_envelope(
     return out
 
 
+@dropping_support
 def extend_pitch(
     base: torch.Tensor,
     pattern: Union[int, List[float], torch.Tensor],
@@ -249,6 +253,7 @@ def extend_pitch(
     return h_freq
 
 
+@dropping_support
 def sinc_impulse_response(cutoff: torch.Tensor, window_size: int = 513, high_pass: bool = False):
     """Create windowed-sinc impulse response for given cutoff frequencies.
 
@@ -288,6 +293,7 @@ def sinc_impulse_response(cutoff: torch.Tensor, window_size: int = 513, high_pas
     return filt
 
 
+@dropping_support
 def frequency_impulse_response(magnitudes):
     """Create filter from desired frequency response
 
@@ -319,6 +325,7 @@ def _overlap_and_add(waveform, stride):
     return buffer
 
 
+@dropping_support
 def filter_waveform(waveform: torch.Tensor, kernels: torch.Tensor, delay_compensation: int = -1):
     """Applies filters along time axis of the given waveform.
 
@@ -404,6 +411,7 @@ def filter_waveform(waveform: torch.Tensor, kernels: torch.Tensor, delay_compens
     return result
 
 
+@dropping_support
 def exp_sigmoid(
     input: torch.Tensor, exponent: float = 10.0, max_value: float = 2.0, threshold: float = 1e-7
 ) -> torch.Tensor:

torchaudio/prototype/functional/_rir.py

@@ -1,5 +1,6 @@
 import math
 from typing import Optional, Tuple, Union
+from torchaudio._internal.module_utils import dropping_support
 
 import torch
 import torchaudio
@@ -176,6 +177,7 @@ def _validate_inputs(
         raise ValueError(f"`mic_array` must be a 2D Tensor with shape (num_channels, 3). Found {mic_array.shape}.")
 
 
+@dropping_support
 def simulate_rir_ism(
     room: torch.Tensor,
     source: torch.Tensor,
@@ -276,6 +278,7 @@ def simulate_rir_ism(
     return rir
 
 
+@dropping_support
 def ray_tracing(
     room: torch.Tensor,
     source: torch.Tensor,

torchaudio/prototype/functional/functional.py

@@ -4,6 +4,7 @@ from typing import Optional
 
 import torch
 from torchaudio.functional.functional import _create_triangular_filterbank
+from torchaudio._internal.module_utils import dropping_support
 
 
 def _hz_to_bark(freqs: float, bark_scale: str = "traunmuller") -> float:
@@ -72,6 +73,7 @@ def _hz_to_octs(freqs, tuning=0.0, bins_per_octave=12):
     return torch.log2(freqs / (a440 / 16))
 
 
+@dropping_support
 def barkscale_fbanks(
     n_freqs: int,
     f_min: float,
@@ -129,6 +131,7 @@ def barkscale_fbanks(
     return fb
 
 
+@dropping_support
 def chroma_filterbank(
     sample_rate: int,
     n_freqs: int,

torchaudio/prototype/models/__init__.py

@@ -1,3 +1,4 @@
+from torchaudio._internal.module_utils import dropping_const_support
 from ._conformer_wav2vec2 import (
     conformer_wav2vec2_base,
     conformer_wav2vec2_model,
@@ -10,7 +11,9 @@ from ._emformer_hubert import emformer_hubert_base, emformer_hubert_model
 from .conv_emformer import ConvEmformer
 from .hifi_gan import hifigan_vocoder, hifigan_vocoder_v1, hifigan_vocoder_v2, hifigan_vocoder_v3, HiFiGANVocoder
 from .rnnt import conformer_rnnt_base, conformer_rnnt_biasing, conformer_rnnt_biasing_base, conformer_rnnt_model
-from .rnnt_decoder import Hypothesis, RNNTBeamSearchBiasing
+from .rnnt_decoder import Hypothesis as _Hypothesis, RNNTBeamSearchBiasing
+
+Hypothesis = dropping_const_support(_Hypothesis, name="Hypothesis")
 
 __all__ = [
     "conformer_rnnt_base",

torchaudio/prototype/models/_conformer_wav2vec2.py

@@ -7,6 +7,7 @@ from torchaudio.models import Wav2Vec2Model
 from torchaudio.models.conformer import ConformerLayer
 from torchaudio.models.rnnt import _TimeReduction
 from torchaudio.models.wav2vec2 import components
+from torchaudio._internal.module_utils import dropping_class_support, dropping_support
 
 
 def _buffered_arange(max) -> Tensor:
@@ -252,6 +253,7 @@ class ConformerEncoder(Module):
         return self._get_intermediate_outputs(x, mask=masks, num_layers=num_layers)
 
 
+@dropping_class_support
 class ConformerWav2Vec2PretrainModel(Module):
     """Conformer Wav2Vec2 pre-train model for training from scratch.
 
@@ -437,6 +439,7 @@ def _get_conformer_negativer_sampler(
     return NegativeSampler(preprocessor, num_negatives, cross_sample_negatives)
 
 
+@dropping_support
 def conformer_wav2vec2_model(
     extractor_input_dim: int,
     extractor_output_dim: int,
@@ -501,6 +504,7 @@ def conformer_wav2vec2_model(
     return Wav2Vec2Model(feature_extractor, encoder)
 
 
+@dropping_support
 def conformer_wav2vec2_base(
     extractor_input_dim: int = 64,
     extractor_output_dim: int = 256,
@@ -536,6 +540,7 @@ def conformer_wav2vec2_base(
     )
 
 
+@dropping_support
 def conformer_wav2vec2_pretrain_model(
     extractor_input_dim: int,
     extractor_output_dim: int,
@@ -672,6 +677,7 @@ def conformer_wav2vec2_pretrain_model(
     )
 
 
+@dropping_support
 def conformer_wav2vec2_pretrain_base(
     extractor_input_dim: int = 64,
     extractor_output_dim: int = 256,
@@ -733,6 +739,7 @@ def conformer_wav2vec2_pretrain_base(
     )
 
 
+@dropping_support
 def conformer_wav2vec2_pretrain_large(
     extractor_input_dim: int = 64,
     extractor_output_dim: int = 256,

torchaudio/prototype/models/_emformer_hubert.py

@@ -4,6 +4,8 @@ import torch
 from torchaudio.models import Wav2Vec2Model
 from torchaudio.models.emformer import Emformer
 from torchaudio.models.rnnt import _TimeReduction
+from torchaudio._internal.module_utils import dropping_support
+
 
 
 class FeatureEncoder(torch.nn.Module):
@@ -217,6 +219,7 @@ def _get_emformer_encoder(
     return EmformerEncoder(emformer, output_linear, layer_norm)
 
 
+@dropping_support
 def emformer_hubert_model(
     extractor_input_dim: int,
     extractor_output_dim: int,
@@ -292,6 +295,7 @@ def emformer_hubert_model(
     return Wav2Vec2Model(feature_extractor, emformer, aux)
 
 
+@dropping_support
 def emformer_hubert_base(
     extractor_input_dim: int = 80,
     extractor_output_dim: int = 128,

torchaudio/prototype/models/conv_emformer.py

@@ -3,6 +3,8 @@ from typing import List, Optional, Tuple
 
 import torch
 from torchaudio.models.emformer import _EmformerAttention, _EmformerImpl, _get_weight_init_gains
+from torchaudio._internal.module_utils import dropping_class_support, dropping_support
+
 
 
 def _get_activation_module(activation: str) -> torch.nn.Module:
@@ -441,6 +443,7 @@ class _ConvEmformerLayer(torch.nn.Module):
         return output_utterance, output_right_context, output_state, next_m
 
 
+@dropping_class_support
 class ConvEmformer(_EmformerImpl):
     r"""Implements the convolution-augmented streaming transformer architecture introduced in
     *Streaming Transformer Transducer based Speech Recognition Using Non-Causal Convolution*
@@ -476,6 +479,7 @@ class ConvEmformer(_EmformerImpl):
         >>> output, lengths, states = conv_emformer.infer(input, lengths, None)
     """
 
+    @dropping_support
     def __init__(
         self,
         input_dim: int,

torchaudio/prototype/models/hifi_gan.py

@@ -28,8 +28,10 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn import Conv1d, ConvTranspose1d
+from torchaudio._internal.module_utils import dropping_class_support, dropping_support
 
 
+@dropping_class_support
 class HiFiGANVocoder(torch.nn.Module):
     """Generator part of *HiFi GAN* :cite:`NEURIPS2020_c5d73680`.
     Source: https://github.com/jik876/hifi-gan/blob/4769534d45265d52a904b850da5a622601885777/models.py#L75
@@ -246,6 +248,7 @@ def get_padding(kernel_size, dilation=1):
     return int((kernel_size * dilation - dilation) / 2)
 
 
+@dropping_support
 def hifigan_vocoder(
     in_channels: int,
     upsample_rates: Tuple[int, ...],
@@ -282,6 +285,7 @@ def hifigan_vocoder(
     )
 
 
+@dropping_support
 def hifigan_vocoder_v1() -> HiFiGANVocoder:
     r"""Builds HiFiGAN Vocoder with V1 architecture :cite:`NEURIPS2020_c5d73680`.
 
@@ -300,6 +304,7 @@ def hifigan_vocoder_v1() -> HiFiGANVocoder:
     )
 
 
+@dropping_support
 def hifigan_vocoder_v2() -> HiFiGANVocoder:
     r"""Builds HiFiGAN Vocoder with V2 architecture :cite:`NEURIPS2020_c5d73680`.
 
@@ -318,6 +323,7 @@ def hifigan_vocoder_v2() -> HiFiGANVocoder:
     )
 
 
+@dropping_support
 def hifigan_vocoder_v3() -> HiFiGANVocoder:
     r"""Builds HiFiGAN Vocoder with V3 architecture :cite:`NEURIPS2020_c5d73680`.
 

torchaudio/prototype/models/rnnt.py

@@ -5,6 +5,8 @@ import torch
 from torchaudio.models import Conformer, RNNT
 from torchaudio.models.rnnt import _Joiner, _Predictor, _TimeReduction, _Transcriber
 
+from torchaudio._internal.module_utils import dropping_support
+
 
 TrieNode = Tuple[Dict[int, "TrieNode"], int, Optional[Tuple[int, int]]]
 
@@ -472,6 +474,7 @@ class RNNTBiasing(RNNT):
         return output, source_lengths, jointer_activation
 
 
+@dropping_support
 def conformer_rnnt_model(
     *,
     input_dim: int,
@@ -544,6 +547,7 @@ def conformer_rnnt_model(
     return RNNT(encoder, predictor, joiner)
 
 
+@dropping_support
 def conformer_rnnt_base() -> RNNT:
     r"""Builds basic version of Conformer RNN-T model.
 
@@ -572,6 +576,7 @@ def conformer_rnnt_base() -> RNNT:
     )
 
 
+@dropping_support
 def conformer_rnnt_biasing(
     *,
     input_dim: int,
@@ -677,6 +682,7 @@ def conformer_rnnt_biasing(
     )
 
 
+@dropping_support
 def conformer_rnnt_biasing_base(charlist=None, biasing=True) -> RNNT:
     r"""Builds basic version of Conformer RNN-T model with TCPGen.
 

torchaudio/prototype/models/rnnt_decoder.py

@@ -4,6 +4,8 @@ import torch
 from torchaudio.models import RNNT
 from torchaudio.prototype.models.rnnt import TrieNode
 
+from torchaudio._internal.module_utils import dropping_class_support
+
 __all__ = ["Hypothesis", "RNNTBeamSearchBiasing"]
 
 
@@ -80,6 +82,7 @@ def _remove_hypo(hypo: Hypothesis, hypo_list: List[Hypothesis]) -> None:
             break
 
 
+@dropping_class_support
 class RNNTBeamSearchBiasing(torch.nn.Module):
     r"""Beam search decoder for RNN-T model with biasing support.
 

torchaudio/prototype/pipelines/__init__.py

@@ -1,6 +1,15 @@
 from ._vggish import VGGISH, VGGishBundle
-from .hifigan_pipeline import HIFIGAN_VOCODER_V3_LJSPEECH, HiFiGANVocoderBundle
-from .rnnt_pipeline import EMFORMER_RNNT_BASE_MUSTC, EMFORMER_RNNT_BASE_TEDLIUM3
+from .hifigan_pipeline import HIFIGAN_VOCODER_V3_LJSPEECH as _HIFIGAN_VOCODER_V3_LJSPEECH, HiFiGANVocoderBundle
+from .rnnt_pipeline import (
+    EMFORMER_RNNT_BASE_MUSTC as _EMFORMER_RNNT_BASE_MUSTC,
+    EMFORMER_RNNT_BASE_TEDLIUM3 as _EMFORMER_RNNT_BASE_TEDLIUM3
+)
+from torchaudio._internal.module_utils import dropping_const_support
+
+EMFORMER_RNNT_BASE_MUSTC = dropping_const_support(_EMFORMER_RNNT_BASE_MUSTC)
+EMFORMER_RNNT_BASE_TEDLIUM3 = dropping_const_support(_EMFORMER_RNNT_BASE_TEDLIUM3)
+HIFIGAN_VOCODER_V3_LJSPEECH = dropping_const_support(_HIFIGAN_VOCODER_V3_LJSPEECH)
+
 
 __all__ = [
     "EMFORMER_RNNT_BASE_MUSTC",

torchaudio/prototype/pipelines/_vggish/__init__.py

@@ -1,3 +1,7 @@
-from ._vggish_pipeline import VGGISH, VGGishBundle
+from ._vggish_pipeline import VGGISH as _VGGISH, VGGishBundle
+from torchaudio._internal.module_utils import dropping_const_support
+
+
+VGGISH = dropping_const_support(_VGGISH, "VGGISH")
 
 __all__ = ["VGGISH", "VGGishBundle"]

torchaudio/prototype/pipelines/_vggish/_vggish_impl.py

@@ -18,6 +18,8 @@ import math
 
 import torch
 
+from torchaudio._internal.module_utils import dropping_class_support
+
 
 _MEL_BREAK_FREQUENCY_HERTZ = 700.0
 _MEL_HIGH_FREQUENCY_Q = 1127.0
@@ -191,6 +193,7 @@ def _waveform_to_examples(data):
     return log_mel_examples.unsqueeze(1)
 
 
+@dropping_class_support
 class VGGish(torch.nn.Module):
     """Implementation of VGGish model :cite:`45611`."""
 
@@ -215,7 +218,7 @@ class VGGish(torch.nn.Module):
 
         return self.embedding_network(x)
 
-
+@dropping_class_support
 class VGGishInputProcessor:
     """Converts raw waveforms to batches of examples to use as inputs to VGGish."""
 

torchaudio/prototype/pipelines/_vggish/_vggish_pipeline.py

@@ -1,8 +1,8 @@
 from dataclasses import dataclass
 from typing import Callable, Dict
 
-import torch
-import torchaudio
+from torchaudio._internal.module_utils import dropping_class_support
+
 
 from ._vggish_impl import _SAMPLE_RATE, VGGish as _VGGish, VGGishInputProcessor as _VGGishInputProcessor
 
@@ -12,6 +12,7 @@ def _get_state_dict():
     return torch.load(path)
 
 
+@dropping_class_support
 @dataclass
 class VGGishBundle:
     """VGGish :cite:`45611` inference pipeline ported from

torchaudio/prototype/pipelines/hifigan_pipeline.py

@@ -9,7 +9,10 @@ from torchaudio._internal import load_state_dict_from_url
 from torchaudio.prototype.models.hifi_gan import hifigan_vocoder, HiFiGANVocoder
 from torchaudio.transforms import MelSpectrogram
 
+from torchaudio._internal.module_utils import dropping_support, dropping_class_support
 
+
+@dropping_class_support
 @dataclass
 class HiFiGANVocoderBundle:
     """Data class that bundles associated information to use pretrained
@@ -82,6 +85,7 @@ class HiFiGANVocoderBundle:
         state_dict = load_state_dict_from_url(url, **dl_kwargs)
         return state_dict
 
+    @dropping_support
     def get_vocoder(self, *, dl_kwargs=None) -> HiFiGANVocoder:
         """Construct the HiFiGAN Generator model, which can be used a vocoder, and load the pretrained weight.
 
@@ -99,6 +103,7 @@ class HiFiGANVocoderBundle:
         model.eval()
         return model
 
+    @dropping_support
     def get_mel_transform(self) -> Module:
         """Construct an object which transforms waveforms into mel spectrograms."""
         return _HiFiGANMelSpectrogram(

torchaudio/prototype/transforms/_transforms.py

@@ -3,8 +3,9 @@ from typing import Callable, Optional
 import torch
 from torchaudio.prototype.functional import barkscale_fbanks, chroma_filterbank
 from torchaudio.transforms import Spectrogram
+from torchaudio._internal.module_utils import dropping_support, dropping_class_support
 
-
+@dropping_class_support
 class BarkScale(torch.nn.Module):
     r"""Turn a normal STFT into a bark frequency STFT with triangular filter banks.
 
@@ -72,6 +73,7 @@ class BarkScale(torch.nn.Module):
         return bark_specgram
 
 
+@dropping_class_support
 class InverseBarkScale(torch.nn.Module):
     r"""Estimate a STFT in normal frequency domain from bark frequency domain.
 
@@ -188,6 +190,7 @@ class InverseBarkScale(torch.nn.Module):
         return specgram
 
 
+@dropping_class_support
 class BarkSpectrogram(torch.nn.Module):
     r"""Create BarkSpectrogram for a raw audio signal.
 
@@ -297,6 +300,7 @@ class BarkSpectrogram(torch.nn.Module):
         return bark_specgram
 
 
+@dropping_class_support
 class ChromaScale(torch.nn.Module):
     r"""Converts spectrogram to chromagram.
 
@@ -356,6 +360,7 @@ class ChromaScale(torch.nn.Module):
         return torch.matmul(x.transpose(-1, -2), self.fb).transpose(-1, -2)
 
 
+@dropping_class_support
 class ChromaSpectrogram(torch.nn.Module):
     r"""Generates chromagram for audio signal.
 

torchaudio/sox_effects/sox_effects.py

@@ -3,7 +3,7 @@ from typing import List, Optional, Tuple
 
 import torch
 import torchaudio
-from torchaudio._internal.module_utils import deprecated
+from torchaudio._internal.module_utils import deprecated, dropping_support
 from torchaudio.utils.sox_utils import list_effects
 
 
@@ -39,6 +39,7 @@ def shutdown_sox_effects():
     pass
 
 
+@dropping_support
 def effect_names() -> List[str]:
     """Gets list of valid sox effect names
 
@@ -52,6 +53,7 @@ def effect_names() -> List[str]:
     return list(list_effects().keys())
 
 
+@dropping_support
 def apply_effects_tensor(
     tensor: torch.Tensor,
     sample_rate: int,
@@ -156,6 +158,7 @@ def apply_effects_tensor(
     return sox_ext.apply_effects_tensor(tensor, sample_rate, effects, channels_first)
 
 
+@dropping_support
 def apply_effects_file(
     path: str,
     effects: List[List[str]],

torchaudio/transforms/__init__.py

@@ -1,3 +1,4 @@
+from torchaudio._internal.module_utils import dropping_class_support
 from ._multi_channel import MVDR, PSD, RTFMVDR, SoudenMVDR
 from ._transforms import (
     AddNoise,
@@ -21,7 +22,7 @@ from ._transforms import (
     PitchShift,
     Preemphasis,
     Resample,
-    RNNTLoss,
+    RNNTLoss as _RNNTLoss,
     SlidingWindowCmn,
     SpecAugment,
     SpectralCentroid,
@@ -34,6 +35,7 @@ from ._transforms import (
     Vol,
 )
 
+RNNTLoss = dropping_class_support(_RNNTLoss)
 
 __all__ = [
     "AddNoise",

torchaudio/transforms/_transforms.py

@@ -10,6 +10,7 @@ from torch.nn.modules.lazy import LazyModuleMixin
 from torch.nn.parameter import UninitializedParameter
 
 from torchaudio import functional as F
+from torchaudio.functional.functional import _rnnt_loss
 from torchaudio.functional.functional import (
     _apply_sinc_resample_kernel,
     _check_convolve_mode,
@@ -1846,7 +1847,7 @@ class RNNTLoss(torch.nn.Module):
             Tensor: Loss with the reduction option applied. If ``reduction`` is  ``"none"``, then size (batch),
             otherwise scalar.
         """
-        return F.rnnt_loss(
+        return _rnnt_loss(
             logits,
             targets,
             logit_lengths,
@@ -2134,4 +2135,4 @@ class Deemphasis(torch.nn.Module):
         Returns:
             torch.Tensor: De-emphasized waveform, with shape `(..., N)`.
         """
-        return F.deemphasis(waveform, coeff=self.coeff)
+        return F.functional.deemphasis(waveform, coeff=self.coeff)